Description:FIELD OF INVENTION
[0001] The present disclosure relates generally to breast cancer screening, and more particularly to a cost-effective system and method for non-invasive breast cancer detection using thermal imaging and artificial intelligence (AI). The system provides a portable, affordable, and radiation-free approach that enables real-time diagnostic results. Further, the system implements a verification process for ensuring the accuracy of thermal measurements of thermal images to improve accuracy in detection of subtle temperature variations that may indicate the presence of cancerous tissue. Due to its high accessibility, the system is well-suited for use in rural areas, remote healthcare centers, and mobile screening units where conventional approaches may be limited or unavailable.
BACKGROUND
[0002] Breast cancer remains one of the most prevalent forms of cancer worldwide, affecting millions of women each year. Early detection plays a crucial role in improving survival rates and treatment outcomes. Traditional screening methods for breast cancer include mammography, magnetic resonance imaging (MRI), and ultrasound. Traditional screening methods have been widely used in healthcare settings for decades and have contributed significantly to the early detection of breast abnormalities. Mammography, in particular, has been the gold standard for breast cancer screening, utilizing X-rays to create detailed images of breast tissue. Magnetic resonance imaging (MRI) employs powerful magnets and radio waves to generate high-resolution images, while ultrasound uses sound waves to produce real-time images of breast tissue.
[0003] Despite the effectiveness of these traditional screening methods, they come with several limitations that hinder their widespread accessibility and frequent use. Mammography, while widely available, exposes patients to ionizing radiation, which can be a concern for repeated screenings. Additionally, mammograms may be less effective in detecting abnormalities in dense breast tissue, which is common in younger women. MRI, although highly sensitive, is expensive and requires specialized equipment, making it less accessible in resource-limited settings. Ultrasound, while non-invasive and radiation-free, is highly operator-dependent and may miss some types of breast abnormalities. These limitations have prompted researchers and healthcare professionals to explore alternative screening methods that can overcome these challenges.
[0004] The high cost and limited accessibility of traditional screening methods pose significant barriers to regular breast cancer screening, particularly in low-resource settings and rural areas. Many women, especially those in developing countries or remote locations, lack access to specialized medical facilities equipped with mammography machines or MRI scanners. This disparity in healthcare access contributes to delayed diagnoses and poorer outcomes for breast cancer patients in these regions. Furthermore, the discomfort and anxiety associated with some screening procedures, such as mammography, can deter women from undergoing regular check-ups, potentially leading to missed opportunities for early detection.
[0005] In recent years, thermal imaging has emerged as a promising alternative for breast cancer screening. However, existing thermal imaging solutions, such as those based on FLIR (Forward-Looking Infrared) systems, are often prohibitively expensive and require powerful processing units, limiting their widespread adoption.
[0006] Artificial intelligence (AI) has been increasingly applied to enhance the accuracy and efficiency of breast cancer screening methods. AI-based thermography models have shown promise in analyzing thermal images and identifying potential abnormalities. However, these models face challenges such as high false-positive rates and dependency on expensive thermal cameras. Moreover, mobile AI-based thermal imaging solutions, while promising, remain largely in the research phase and often require high-end sensors, making them impractical for widespread implementation in clinical settings. The limitations of current AI-based approaches highlight the need for more affordable and accessible solutions that can leverage the benefits of thermal imaging and artificial intelligence for breast cancer screening.
[0007] Therefore, there is a need to overcome the problems discussed above in breast cancer screening.
OBJECTIVES
[0008] The primary objective of the present disclosure is to provide a system for breast cancer screening using thermal imaging and artificial intelligence (AI).
[0009] Another objective of the present disclosure is to offer a portable edge device for local data processing in breast cancer screening, mass deployment in rural healthcare, telemedicine, and mobile clinics, enabling low-cost breast cancer screening.
[0010] Yet another objective of the present disclosure is to provide an image preprocessing technique for optimizing thermal images for breast cancer screening.
[0011] Yet another objective of the present disclosure is to provide a functionality for verifying temperature values of the thermal images.
[0012] Yet another objective of the present disclosure is to offer a breast cancer screening solution that operates without requiring internet connectivity or cloud-based processing, making it suitable for use in remote or rural areas with limited network infrastructure.
[0013] A still further objective of the present disclosure is to provide a cost-effective alternative to traditional mammography screening, potentially increasing access to breast cancer early detection in resource-limited settings.
[0014] Another objective of the present disclosure is to develop a non-invasive and radiation-free method for breast cancer screening, allowing for more frequent examinations without the risks associated with ionizing radiation.
[0015] An additional objective of the present disclosure is to create a user-friendly system that can be operated by healthcare workers with minimal specialized training, thereby expanding the reach of breast cancer screening programs.
[0016] A further objective of the present disclosure is to enable real-time analysis and immediate display of screening results.
SUMMARY
[0017] The present disclosure provides a technical solution for early detection of breast cancer using a thermal imaging and artificial intelligence, addressing key challenges in breast cancer screening. The present disclosure provides a system and a method for breast screening. By utilizing affordable sensors and widely available computing hardware, the system improves accessibility, especially in low-resource settings and rural areas, while offering a completely radiation-free and non-contact method of breast examination. The integration of edge computing and deep learning models enables real-time analysis, providing immediate results to healthcare providers and patients. A significant aspect of this solution is the verification of thermal image values using single-point temperature measurements from infrared sensors, which ensures the accuracy and reliability of the thermal data. This verification process helps to detect any potential sensor malfunctions or imaging artifacts, thereby improving the overall quality and trustworthiness of the breast cancer screening results. The portable design suits mobile clinics and telemedicine applications, while AI-based image analysis reduces subjectivity in interpretation. The system is cost-effective, scalable alternative to expensive medical imaging equipment, potentially reducing healthcare costs and adapting to various clinical settings. The non-contact nature improves patient comfort and compliance, potentially increasing screening participation rates, while the collection of standardized thermal imaging data enables the creation of large datasets for further research and improvement of AI models in breast cancer detection. This technical solution represents a significant advancement in breast cancer screening technology, offering the potential to improve early detection rates, reduce healthcare disparities, and ultimately save lives through more accessible and frequent breast examinations.
[0018] According to one aspect of the present disclosure, a system for breast cancer screening is provided. The system comprises at least one thermal imaging camera, at least two temperature sensors, and an edge computing unit. The thermal imaging camera is configured for capturing one or more thermal images of a breast of a subject. The captured thermal images comprise first pixel values. The temperature sensors are configured for measuring temperature values at one or more target locations of the breast of the subject. The edge computing unit comprises a memory storing a set of instructions and a processing unit. The thermal imaging camera and the temperature sensors are positioned in the edge computing unit or communicatively coupled to the edge computing unit. The processing unit is configured to execute the set of instructions to: obtain the thermal images of the breast of the subject; obtain temperature data corresponding to the one or more target locations of the breast of the subject; verify temperature values of the thermal images based on temperature data corresponding to the one or more locations of the breast of the subject; preprocess verified thermal images; extract, using one or more deep learning convolution models deployed in the edge computing unit, one or more temperature based features from the preprocessed thermal images; generate, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images; and classify, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score. The temperature sensors are non-contact infrared thermal sensors.
[0019] The processing unit is further configured to display, on a display unit, at least one of the thermal images of the breast, classification result associated with the thermal images, and warning message if abnormality is detected, or combination thereof.
[0020] The processing unit is configured to verify temperature values of the thermal images by selecting one or more regions of interest in the thermal images that correspond to the one or more target locations of the breast; averaging or extracting temperature values from the one or more regions of interest in the thermal images; comparing temperature values from the one or more regions of interest in the thermal images with corresponding temperature points in the one or more target locations of the breast; and accepting the thermal images for preprocessing if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are matching.
[0021] The processing unit is further configured to instruct to capture new thermal images of the breast if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are not matching.
[0022] The processing unit is configured to preprocess verified thermal images by: removing noise from the thermal images by applying one or more filters; normalizing pixel intensity values of the thermal images; identifying and isolating breast regions by removing background areas; enhancing contrast of identified and isolated breast regions to improve visibility of subtle temperature differences; resizing enhanced thermal images comprising first pixel values into thermal images comprising second pixel values; and optionally performing color mapping or alignment if multiple views are used.
[0023] The processing unit is configured to extract the one or more temperature-based features from the preprocessed thermal images by: extracting mean, maximum, minimum, median, standard deviation of temperatures within the one or more regions of interest; extracting spatial features by performing hotspot analysis and symmetry analysis between left and right breast regions; extracting, using grey-level co-occurrence matrix (GLCM), textural features to assess contrast, correlation, and entropy; and extracting histogram features by analysing temperature distribution patterns across the preprocessed thermal images.
[0024] The processing unit is further configured to generate one or more features vectors based on the extracted one or more temperature-based features, for providing as an input to the one or more deep learning convolution models.
[0025] The one or more deep learning convolution models are trained by obtaining a plurality of thermal images of breast and temperature data at one or more target locations of corresponding breast tissue; verifying temperature values of one or more regions of interest in the thermal images to the temperature data at the corresponding target locations in the breast; preprocessing verified thermal images; labeling the preprocessed thermal images as cancerous or noncancerous; extracting one or more temperature based features from the preprocessed plurality of thermal images; and training the one or more deep learning convolution models with extracted features and associated diagnostic labels and preprocessed thermal images
[0026] According to one aspect of the present disclosure, a method for breast cancer screening is provided. The method comprises: capturing, using at least one thermal camera, one or more thermal images of a breast of a subject, wherein captured thermal images comprise first pixel values; measuring, using at least two temperature, temperature values at one or more target locations of the breast of the subject; verifying, using an edge computing unit, temperature values of the thermal images based on temperature data corresponding to the one or more target locations of the breast of the subject; preprocessing, using the edge computing unit, verified thermal images; extracting, using one or more deep learning convolution models deployed in the edge computing unit, one or more temperature based features from the preprocessed thermal images; generating, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images; and classifying, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score.
[0027] The foregoing paragraphs have been provided by way of general introduction and are not intended to limit the scope of the following claims. The described embodiments, together with further advantages, will be best understood by reference to the following detailed description taken in conjunction with the accompanying drawings.
BRIEF DESCRIPTION OF DRAWINGS
[0028] FIG. 1 is a block diagram illustrating a system for breast cancer screening in accordance with the present disclosure.
[0029] FIG. 2 is a block diagram illustrating one or more modules of an edge computing unit of FIG. 1 in accordance with the present disclosure.
[0030] FIGS. 3A-3B are flow charts illustrating a method of breast cancer screening in accordance with the present disclosure.
[0031] FIG. 4 is a flow chart illustrating a method of training a deep learning convolution neural network model in accordance with the present disclosure.
[0032] FIG. 5 is a block diagram illustrating an edge computing unit in accordance with the present disclosure.
DETAILED DESCRIPTION OF THE PRESENT DISCLOSURE
[0033] Aspects of the present invention are best understood by reference to the description set forth herein. All the aspects described herein will be better appreciated and understood when considered in conjunction with the following descriptions. It should be understood, however, that the following descriptions, while indicating preferred aspects and numerous specific details thereof, are given by way of illustration only and should not be treated as limitations. Changes and modifications may be made within the scope herein without departing from the spirit and scope thereof, and the present invention herein includes all such modifications.
[0034] As mentioned above, there is a need for a technical solution to solve aforementioned technical problems in breast cancer screening. The present disclosure provides a system and a method for breast cancer screening that combines thermal imaging technology, temperature sensing, and artificial intelligence-driven analysis. The system verifies thermal image data against precise point temperature measurements, ensuring the accuracy and reliability of the screening process. The verification step helps to mitigate potential errors arising from environmental factors or sensor discrepancies, thereby enhancing the overall robustness of the screening method.
[0035] The system's design prioritizes portability, cost-effectiveness, and ease of use, making it suitable for deployment in a wide range of healthcare settings, including resource-limited areas and mobile screening units. By leveraging affordable, off-the-shelf components and open-source software platforms, the system offers a significant reduction in cost compared to traditional breast cancer screening methods, without compromising on diagnostic capability.
[0036] Furthermore, the non-invasive and radiation-free nature of this thermal imaging-based approach allows for more frequent screenings, potentially enabling earlier detection of breast cancer. The integration of artificial intelligence not only enhances the accuracy of detection but also reduces the reliance on specialized radiologists for image interpretation, thereby addressing the shortage of skilled personnel in many healthcare systems.
[0037] As used herein, several terms are defined below:
[0038] The term “thermal imaging camera” as used herein generally refers to a thermal imaging device capable of capturing infrared radiation emitted by an object and converting it into a visual representation.
[0039] The term “thermal images” as used herein generally refers to thermal images captured by a thermal imaging camera, which represent temperature distributions of the captured scene.
[0040] The term “temperature sensor” as used herein generally refers to a device or apparatus, instrument capable of measuring temperature without physical contact, such as an infrared thermometer or a non-contact temperature probe.
[0041] The term “an edge computing unit” as used herein generally refers to a computing device capable of performing data processing and analysis at or near the source of data generation, without relying on cloud computing or internet connectivity.
[0042] The term “pixel values” as used herein generally refers to the numerical representation of temperature or intensity at each point (pixel) in a thermal image.
[0043] The term “deep learning convolution models” as used herein generally refers to artificial neural network architectures, such as convolutional neural networks (CNNs), configured to automatically learn and extract relevant features from image data for tasks such as classification or anomaly detection.
[0044] The term “verifying” as used herein generally refers to the process of validating and cross-checking the temperature values obtained from the thermal imaging camera against the precise temperature measurements taken by the infrared temperature sensors at specific target locations on the breast.
[0045] The term “target location in the breast” as used herein generally refers to specific anatomical points or areas on the breast that are selected for precise temperature measurement using temperature sensors.
[0046] The term “region of interest in thermal images” as used herein generally refers to specific areas within the captured thermal images.
[0047] The term “preprocessing” as used herein generally refers to a series of image processing techniques applied to raw thermal images to enhance their quality, standardize their format, and prepare them for analysis by deep learning models.
[0048] Referring now to the drawings, and more particularly to FIGS. 1 through 5, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments.
[0049] FIG. 1 is a block diagram illustrating a system 100 for breast cancer screening in accordance with the present disclosure. The system 100 includes an edge computing unit 102, a thermal imaging camera 108, a first temperature sensor 110, and a second temperature sensor 112. The thermal imaging camera 108, the first temperature sensor 110, and the second temperature sensor 112 are communicatively connected with the edge computing unit 102. For instance, the thermal imaging camera 108, the first temperature sensor 110, and the second temperature sensor 112 are communicatively connected with the edge computing unit 102, using a communication interface. The connection may be established either via wired means or wirelessly. In some embodiments, the wired connection may utilize protocols such as USB (Universal Serial Bus), SPI (Serial Peripheral Interface), or I2C (Inter-Integrated Circuit) for data transmission. These wired connections can provide stable, high-speed data transfer with minimal latency. In other embodiments, wireless communication may be employed, utilizing protocols such as Wi-Fi, Bluetooth Low Energy (BLE), or ZigBee. Wireless connections offer flexibility in device placement and can contribute to the overall portability of the system 100. In some embodiments, the system 100 includes a data acquisition device configured for collecting the thermal images and temperature data, respectively, from the thermal imaging camera 108, the first temperature sensor 110, and the second temperature sensor 112 and transmitting the thermal images and temperature data to the edge computing unit 102. The data acquisition device connected with the edge computing unit 102 via wired means or wirelessly. In one exemplary embodiment, the data acquisition device is a microcontroller board.
[0050] The thermal imaging camera 108 is configured for capturing one or more thermal images of a breast of a subject. It is to be noted that thermal images of the breast refer both left side and right side of the breast. In one exemplary embodiment, the thermal imaging camera 108 is a low resolution thermal imaging camera. The thermal imaging camera 108 may have a resolution of 32×24 pixels. In some embodiments, the thermal imaging camera 108 have a resolution below 32×24 pixels. In some embodiments, the thermal imaging camera 108 has a resolution above 32×24 pixels. In some embodiments, the field of view (FoV) of the thermal imaging camera 108 is 55° × 35° (standard) or 110° × 75° (wide-angle version). The accuracy includes ±1.5°C absolute, ±0.1°C relative.
[0051] It is to be noted that more than one thermal imaging camera may be used. For instance, multiple thermal imaging cameras are positioned at different angles to increase coverage of the breast tissue, capturing a more comprehensive thermal map. The multi-angle approach can help detect anomalies that might be less apparent from a single viewpoint, potentially improving the detection of tumors located in areas that are challenging to image with a single camera.
[0052] In some embodiments, the thermal imaging camera 108 comprises a lens system, an infrared detector and a signal processing unit. The lens system is configured for focusing the infrared radiation emitted from a right side breast and a left side of breast of the subject, onto the infrared detector. The infrared detector is configured for capturing infrared radiation emitted from the right side breast and the left side of breast of the subject and converting captured infrared radiation into an electrical signal. In one exemplary embodiment, the infrared detector is a microbolometer. The signal processing unit is configured for converting the electrical signals into thermal images, where different temperatures are represented by different colors. In one exemplary embodiment, warmer areas are typically represented in red or yellow, while cooler areas are in blue or green. In some embodiments, the captured thermal images comprise first pixel values. In one exemplary embodiment, the first pixel values comprise 32×24 pixels.
[0053] The temperature sensors 110, 112 are configured for measuring temperature at one or more target locations in the breast of the subject. In some embodiments, the temperature sensors 110, 112 are non-contact infrared (IR) temperature sensors. The temperature sensors 110, 112 are positioned to capture temperature data from clinically significant areas of the breast. In one exemplary embodiment, the target locations include upper outer quadrant of each breast, as this area statistically has the highest incidence of breast cancer, areola and nipple area, which can exhibit temperature changes in cases of underlying tumors, inframammary fold, where temperature differences may indicate abnormalities in the lower portion of the breast, symmetrical points on both breasts to allow for comparative analysis or any areas of concern identified through physical examination. In some embodiments, the temperature sensors 110, 112 are positioned to capture temperature data from all anatomical points in the breast, thereby covering the entire breast area. The use of two temperature sensors allows for simultaneous measurement of corresponding points on both breasts, enabling direct comparison of temperature symmetry. The precise point measurements serve as reference points for verifying the temperature values of thermal images. In one exemplary embodiment, the temperature sensors 110, 112 comprise a lens to collect and focus IR radiation emitted by the subject; a detector, often a thermopile, to convert the focused IR radiation into an electrical signal; and a signal processing unit to convert the electrical signal to temperature readings. This non-invasive approach enhances patient comfort and eliminates the risk of cross-contamination between subjects. To ensure accuracy, the temperature sensors 110, 112 include an internal ambient temperature compensation mechanism, thereby adjusting for variations in the sensor's own temperature, which could otherwise affect the accuracy of the body surface temperature readings. It is to be noted that more than two temperature sensors may also be used. It is to be noted that more than two temperature sensors may also be used to further enhance the accuracy and comprehensiveness of the breast cancer screening system. Incorporating additional temperature sensors offers several advantages, including increased spatial resolution for detecting subtle temperature variations, enhanced verification capability for improved reliability, better coverage of complex breast geometries, and improved redundancy and fault tolerance. The redundancy provided by multiple sensors ensures screening continuity even if some sensors malfunction.
[0054] The thermal imaging camera 108, and the temperature sensors 110, 112 are configured to transmit the thermal images, and temperature data, respectively, to the edge computing unit 102, directly or indirectly. In this context, "directly" refers to a configuration where the sensors and camera are connected to the edge computing unit 102 through direct wired or wireless interfaces, such as USB, Ethernet, Wi-Fi, or Bluetooth, without any intermediate processing or storage devices. The direct connection allows for immediate transmission of raw data to the edge computing unit 102 for processing. On the other hand, "indirectly" implies the use of an intermediary device or system, such as the Arduino Uno microcontroller, which acts as a data acquisition and preprocessing unit. In the indirect configuration, the thermal imaging camera 108 and temperature sensors 110, 112 transmit their data to the intermediary device, which may perform initial data formatting, synchronization, or basic processing before forwarding the collected data to the edge computing unit 102. The indirect approach can offer benefits such as data buffering, sensor fusion, and reduced computational load on the main processing unit.
[0055] The edge computing unit 102 comprises a memory 104 storing a set of instructions and a processing unit 106. In some embodiments, the memory 104 is a non-volatile storage medium configured to store the set of instructions, data, and deep learning models required for the breast cancer screening process. In some embodiments, the memory 104 comprises a combination of high-speed RAM (random access memory) for temporary data storage and processing, and flash memory or solid-state drive (SSD) for long-term storage of instructions, models, and the like. The processing unit 106 is a computational engine optimized for edge computing and AI inference. In some embodiments, the processing unit 106 comprises anyone of a central processing unit (CPU), a graphics processing unit (GPU), a neural processing unit (NPU), a digital signal processor (DSP), or combinations thereof. It is to be noted that edge computing unit 102 is any low-cost computing device capable of running preprocessing algorithms and AI models. For instance, the edge computing unit 102 is a laptop, tablet, single board computers, any edge computing unit with processing capabilities. It is to be noted that the examples are given for illustrative purposes and do not restrict the scope of the present disclosure. The edge computing unit 102 is configured to run any one of Linux, Android, or a lightweight real-time operating system (RTOS), and is equipped with sufficient processing resources to perform local data acquisition, signal processing, and inference operations without reliance on cloud-based computing infrastructure. The edge computing unit 102 is configured to be flexible and can support various programming languages for image processing, data analysis, and machine learning tasks. The programming languages may include, but are not limited to, python, C++ MATLAB, and Java. The choice of programming language can be tailored to the specific requirements of the deployment scenario, considering factors such as processing speed, memory usage, available libraries, and ease of integration with other system components.
[0056] The edge computing unit 102 is configured to obtain the thermal images of the breast of the subject; obtain temperature data corresponding to the one or more target locations of the breast of the subject; verify temperature values of the thermal images based on temperature data corresponding to the one or more locations of the breast of the subject; preprocess verified thermal images; extract, using one or more deep learning convolution models deployed in the edge computing unit 102, one or more temperature based features from the preprocessed thermal images; generate, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images; and classify, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score. If the obtained thermal images are not acceptable, the edge computing unit 102 is further configured to instruct to capture new thermal images of the breast.
[0057] The edge computing unit 102 further includes a display unit 114 that is configured to display at least one of the thermal images of the breast, the temperature data, classification result associated with the thermal images, and warning message if abnormality is detected, or combination thereof. In some embodiments, the display unit 114 is selected from a group comprising of a liquid crystal display (LCD), an organic light-emitting diode (OLED) display, an e-paper display, a touchscreen display, a light-emitting diode (LED) matrix, and a seven-segment display. It is to be noted that any display unit compatible with the chosen edge computing platform and meeting the basic requirements for displaying the breast cancer screening information can be utilized in the edge computing unit 102. The selection of the display unit 114 is not limited to the examples provided above.
[0058] In some embodiments, the system 100 includes a cooling fan for stabilizing temperature of the temperature sensors 110, 112 for improving accuracy of the temperature sensors 110, 112. The fan may create a stable thermal environment around the the temperature sensors 110, 112, reducing thermal drift and dissipating heat generated during operation. By minimizing thermal noise and ensuring consistent operating temperatures, the fan enables the temperature sensors 110, 112 to detect even subtle temperature variations in breast tissue more effectively.
[0059] FIG. 2 is a block diagram illustrating one or more modules of the edge computing unit 102 of FIG. 1 in accordance with the present disclosure. The edge computing unit 102 comprises a database 200, a thermal image obtaining module 202, a temperature data obtaining module 204, a verification module 206, a preprocessing module 208, a feature extraction module 210, a score generating module 212, a classification module 214, and a display module 216. It should be understood that this modular structure is flexible and adaptable to various implementation requirements and future enhancements. One or more additional modules may be incorporated into the system 100 to expand its functionality, such as a data encryption module for enhanced security or a telemedicine module for remote consultation capabilities. Conversely, existing modules may be combined to optimize processing efficiency; for instance, the feature extraction module 210 and score generating module 212 could potentially be merged into a single analysis module. Similarly, modules may be subdivided for more granular control or to accommodate specific hardware configurations; the preprocessing module 208, for example, could be split into separate noise reduction and image enhancement modules. This modular approach allows for scalability and customization of the edge computing unit 102, ensuring it can be adapted to meet evolving technological advancements and diverse healthcare needs while maintaining its core breast cancer screening functionality.
[0060] The database 200 stores thermal images, temperature data, deep learning model parameters for quick loading and execution, securely encrypted user profiles, anonymized historical screening records for longitudinal studies; system logs for maintenance and optimization; standardized reference ranges for result interpretation; and a repository of verified software updates. In one exemplary embodiment, the database 200 utilizes SQLite, a self-contained, serverless, and zero-configuration database engine. SQLite is chosen for its small footprint, low resource requirements, and compatibility with various low-power computing platforms. In some embodiments, the edge computing unit 102 may utilize external database for enhanced data management and storage capabilities.
[0061] The thermal image obtaining module 202 is configured to obtain the thermal images of the breast of the subject from the database 200 or the thermal imaging camera 108. The thermal image obtaining module 202 functions as an interface for thermal image acquisition, offering flexibility in image sourcing to accommodate various operational scenarios. In some embodiments, when obtaining images directly from the thermal imaging camera 108, the thermal image obtaining module 202 manages real-time image capture, controlling parameters such as exposure time and frame rate to ensure optimal image quality. The thermal image obtaining module 202 also implements a standardized imaging protocol, guiding the operator to capture consistent views of both breasts for accurate comparative analysis. In cases where images are retrieved from the database 200, the thermal image obtaining module 202 employs data retrieval algorithms to quickly access stored thermal images, which may be useful for follow-up screenings or offline analysis. Additionally, the thermal image obtaining module 202 performs initial quality checks on the obtained images, flagging any that may be suboptimal due to factors like motion blur or improper framing, thereby maintaining the integrity of subsequent analysis steps. By supporting both real-time capture and database retrieval, the thermal image obtaining module 202 enhances the versatility and reliability of the breast cancer screening system 100, allowing it to function effectively in both connected and standalone operational modes.
[0062] The temperature data obtaining module 204 is configured to obtain temperature data corresponding to the one or more target locations of the breast of the subject from the temperature sensors 110, 112 or from the database 200. The temperature data obtaining module 204 function as a central hub for temperature data collection and management. The temperature data obtaining module 204 ensures synchronized temperature data acquisition with thermal imaging, adapts to diverse operational settings, and maintains measurement accuracy, thereby providing critical contextual information for precise interpretation of thermal images and enhancing the overall efficacy of the breast cancer screening process.
[0063] The verification module 206 is configured to verify temperature values of the thermal images based on temperature data corresponding to the one or more locations of the breast of the subject. First, the verification module 206 is configured to select one or more regions of interest in the thermal images that correspond to the one or more target locations of the breast. For instance, consider a thermal image of a left breast captured by the thermal imaging camera 108. One target location where point temperature measured, is the upper outer quadrant of the breast. Then, a pixel area corresponding to the upper outer quadrant is selected. Similarly, the process would be repeated for other key locations, such as the nipple area, and the inframammary fold, ensuring that the entire thermal image accurately represents the breast's temperature distribution. Thereafter, the verification module 206 is configured to perform averaging or extracting temperature values from the one or more regions of interest in the thermal images. In some embodiments, the verification module 206 calculates the average temperature values of the selected pixel area. The averaging process helps to mitigate the effects of any single-pixel anomalies or noise in the thermal image. In some embodiments, the verification module 206 extracts other statistical measures from the selected pixel area, such as the median temperature (to reduce the impact of outliers), the maximum temperature (to identify potential hot spots), or the standard deviation (to assess temperature uniformity within the region). The extracted values provide a more comprehensive representation of the temperature distribution in the selected pixel area.
[0064] The verification module 206 is further configured to compare temperature values from the one or more regions of interest in the thermal images with corresponding temperature points in the one or more target locations of the breast, thereby ensuring the accuracy and reliability of the thermal imaging data. The verification module 206 performs a point-by-point comparison between the processed thermal image data and the precise measurements from temperature sensors 110, 112. For instance, continuing our example, the average temperature of 35.7°C calculated from the 5x5 pixel region in the thermal image (upper outer quadrant of the left breast) is compared with the reading of 35.8°C from the corresponding MLX90614 sensor. The module applies a predefined tolerance threshold, typically ±0.5°C, to determine if the values are sufficiently matched. In this case, the 0.1°C difference falls within the acceptable range, so this region would be considered verified. The verification module 206 repeats this process for all designated regions of interest, such as the nipple area and inframammary fold, creating a comprehensive verification profile for the entire breast. If any comparison falls outside the tolerance range, the verification module 206 flags that region for further investigation or potential recapture. Additionally, the verification module 206 may employ more sophisticated comparison techniques, such as analyzing the temperature gradient or pattern around each point, to ensure not just absolute temperature accuracy but also the correct relative distribution of heat across the breast tissue. The thorough verification process significantly enhances the reliability of the subsequent analysis, reducing the risk of false positives or negatives due to measurement inaccuracies or environmental factors.
[0065] The verification module 206 is further configured to accept the thermal images for preprocessing if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are matching. The acceptance criterion ensures that only high-quality, accurate thermal data proceeds to the next stage of analysis. For example, if all compared regions (such as the upper outer quadrant, nipple area, and inframammary fold) show temperature differences within the predefined tolerance, the entire thermal image is deemed verified and accepted. In one exemplary embodiment, the predefined tolerance is set at ±0.5°C. It is to be noted that the predefined tolerance can be optimized according to the specific requirements of the screening environment, patient characteristics, and clinical objectives. The verification module 206 then passes the verified image to the preprocessing stage, along with a confidence metric indicating the degree of match across all compared points.
[0066] The verification module 206 is further configured to instruct to capture new thermal images of the breast if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are not matching. The fail-safe mechanism is crucial for maintaining data integrity. For instance, if the upper outer quadrant shows a discrepancy between the thermal image and the sensor reading (exceeding the predefined tolerance), the verification module 206 initiates a recapture protocol.
[0067] The cross-validation between two independent measurement techniques, enhancing overall accuracy and reliability. The acceptance criteria based on matching temperatures ensures that only high-quality, consistent thermal data proceeds to the preprocessing stage, maintaining the integrity of subsequent analyses. The multi-step verification process includes detecting and correcting calibration issues in the thermal camera, compensating for environmental factors that might affect measurements, identifying any malfunctioning sensors or image capture problems, ensuring consistency across different screening sessions, and providing a quantitative basis for quality control in the thermal imaging process. By implementing the verification step, the system significantly enhances its ability to detect subtle temperature variations indicative of breast abnormalities, while simultaneously reducing false positives that might arise from measurement inaccuracies or environmental factors. This approach leads to more reliable and reproducible breast cancer screening results, increasing the overall effectiveness and trustworthiness of the thermal imaging-based detection system.
[0068] The preprocessing module 208 is configured to preprocess the verified thermal images. First, the preprocessing module 208 is configured to remove noise from the thermal images by applying one or more filters. The one or more filters include, but not limited to, gaussian filters, median filters, bilateral filters, anisotropic diffusion filters, wavelet-based denoising filters, adaptive local noise reduction filters or combination thereof. The gaussian filters may be applied to smooth out general thermal noise, reducing random fluctuations in pixel values while preserving larger-scale temperature patterns. The median filters may be employed to remove salt-and-pepper noise, which can occur due to sensor imperfections or environmental interference. For preserving edge information while reducing noise, bilateral filters or anisotropic diffusion filters might be utilized, allowing for noise reduction while maintaining important thermal boundaries that could indicate tissue transitions or potential abnormalities. In cases where temporal data is available (if multiple frames are captured), temporal averaging or Kalman filtering can be applied to reduce time-dependent noise. Additionally, wavelet-based denoising techniques may be employed for more sophisticated noise removal, particularly effective at preserving fine thermal details while eliminating noise across different scales. The noise removal process significantly enhances the clarity and accuracy of the thermal images, providing a cleaner dataset for feature extraction and anomaly detection in the breast cancer screening process.
[0069] Thereafter, the preprocessing module 208 is configured to normalize pixel intensity values of the thermal images, thereby standardizing the thermal data for consistent analysis across different images and patients. The normalization process involves scaling the pixel intensity values to a standard range, often 0-1, which helps in mitigating variations due to differences in ambient temperature, sensor sensitivity, or individual physiological factors. The normalization technique includes, but not limited to, min-max normalization, z-score normalization, histogram equalization, adaptive histogram equalization, and reference-based normalization. The normalization approach ensures that subsequent analysis steps, particularly those involving machine learning algorithms, can operate on consistently scaled data, thereby improving the accuracy and reliability of the breast cancer detection process.
[0070] The preprocessing module 208 is further configured to identify and isolating breast regions by removing background areas. The preprocessing module 208 may employ image segmentation techniques to delineate the breast contours from the surrounding thermal background. For instance, the preprocessing module 208 may utilize edge detection algorithms to identify the breast boundaries based on temperature gradients. It is to be noted that any image analysis technique can be used to remove background areas in the thermal image.
[0071] The preprocessing module 208 is further configured to enhance contrast of identified and isolated breast regions to improve visibility of subtle temperature differences. In some embodiments, the preprocessing module 208 starts with histogram equalization, stretching the temperature distribution across the full range of available values, which can reveal subtle variations that might be obscured in the original image. For instance, the histogram equalization includes, but not limited to, adaptive histogram equalization (AHE) or contrast limited adaptive histogram equalization (CLAHE). For even more precise control, the module might implement multi-scale contrast enhancement, applying different levels of enhancement to various spatial frequencies within the image.
[0072] The preprocessing module 208 is further configured to resize enhanced thermal images comprising first pixel values into thermal images comprising second pixel values. In one exemplary embodiment, the first pixel values include 32 × 24 pixels. The first pixel values include 224×224 pixels. In one exemplary embodiment, the resizing step employs interpolation techniques to ensure minimal loss of thermal information during upscaling. The resizing step standardizes the image dimensions and ensuring compatibility with subsequent analysis algorithms, particularly deep learning models.
[0073] The preprocessing module 208 is optionally configured to perform color mapping or alignment if multiple views are used. The color mapping involves assigning specific colors to different temperature ranges, enhancing the visual representation of thermal variations in the breast tissue. Alignment, on the other hand, is critical when multiple views of the breast are captured. The alignment includes frontal and lateral views, or images taken at different time points. The alignment process ensures that these multiple views are spatially registered, allowing for accurate comparison and analysis. The alignment step might involve techniques such as affine transformations or more advanced non-rigid registration methods to account for differences in breast position or shape between views. The module may employ feature-based alignment algorithms, identifying key anatomical landmarks in each image and using these to guide the alignment process. Alternatively, intensity-based registration techniques might be used, which aim to maximize the similarity of temperature patterns across different views. By performing these optional steps, the preprocessing module 208 enhances the interpretability of the thermal images and ensures consistency across multiple views, contributing to more accurate and comprehensive breast cancer screening.
[0074] It is to be noted that the preprocessing steps described above may typically be performed in the sequence presented. However, the order of these steps can be varied or customized based on specific requirements of the breast cancer screening process or characteristics of the thermal imaging data. For instance, in some implementations, noise removal might be performed after breast region isolation to focus denoising efforts specifically on the areas of interest. Alternatively, contrast enhancement could be applied before normalization in cases where highlighting subtle temperature differences is prioritized over standardizing the overall temperature range. The system 100 allows for flexibility in the preprocessing pipeline, enabling optimization of the image preparation process for different clinical scenarios, imaging conditions, or patient-specific factors. This adaptability ensures that the thermal images are processed in the most effective manner for subsequent analysis, regardless of variations in input data quality or specific screening objectives. The preprocessing module 208 may even employ adaptive algorithms that dynamically determine the optimal sequence of preprocessing steps based on the characteristics of each individual thermal image, further enhancing the robustness and versatility of the breast cancer screening system. Due to preprocessing techniques, even low resolution thermal images can also be used for breast cancer screening.
[0075] The feature extraction module 210 is configured to one or more temperature based features from the preprocessed thermal images. The feature extraction module 210 is configured to extract mean, maximum, minimum, median, standard deviation of temperatures within the one or more regions of interest by analyzing the temperature values in the region of interest. For instance, the mean temperature is calculated by averaging all pixel temperature values in the region of interest, providing an overall measure of heat in that region of interest. The maximum temperature is calculated by identifying the highest temperature point, which could indicate a potential hot spot. The minimum temperature is calculated by identifying the lowest temperature point, useful for understanding the temperature range. The median temperature is determined by calculating middle value of all temperatures in the region of interest, less affected by extreme values than the mean. The standard deviation is calculated by measuring the spread of temperature values, indicating how uniform or varied the heat distribution is in the region. The statistical features provide a comprehensive summary of the temperature characteristics in each region of interest, allowing for detailed comparison between different areas of the breast and potentially highlighting anomalies that might be indicative of breast cancer. The feature extraction module 210 calculates the statistical features for each defined region, creating a quantitative profile of the breast's thermal patterns.
[0076] The feature extraction module 210 is configured to extract spatial features by performing hotspot analysis and symmetry analysis between left and right breast regions. In some embodiments, the hotspot analysis involves identifying areas with significantly higher temperatures compared to surrounding tissue; calculating the size, shape, and intensity of the identified hotspots; and determining the number and distribution of hotspots across the breast. In some embodiments, the symmetry analysis includes comparing corresponding regions in the left and right breasts (e.g., upper outer quadrants); calculating temperature differences between the symmetrical regions; and assessing overall thermal patterns and distributions in both breasts. For instance, the feature extraction module 210 may detect a hotspot in the upper outer quadrant of the right breast, measuring 1.5°C warmer than the surrounding tissue and covering an area of approximately 2 cm²; compare the identified area to the corresponding area in the left breast, finding a temperature difference of 1.2°C between the two sides; and analyze the overall symmetry of heat patterns, noting any significant asymmetries that could indicate abnormal tissue growth. The spatial features provide information about the localization and extent of potential abnormalities, leveraging the principle that cancerous growths often create asymmetrical heat patterns or distinct hotspots due to increased metabolic activity and blood flow.
[0077] The feature extraction module 210 is configured to extract, using grey-level co-occurrence matrix (GLCM), textural features to assess contrast, correlation, and entropy. The textural feature analysis process involves analyzing the spatial relationships of temperature values in the thermal image. For instance, contrast is assessed by measuring local variations in the thermal image. High contrast indicates a high degree of temperature differences between neighboring areas. Further, the linear dependency of temperature values on the neighboring pixels is analyzed when high contrast pixels are identified. It shows how correlated a pixel is to its neighbor over the entire image. Further, randomness or complexity of temperature distribution in the thermal image is quantified. Higher entropy suggests more complex thermal patterns. The feature extraction module 210 calculates the textural features for different directions and distances between pixel pairs, typically at 0°, 45°, 90°, and 135° angles and for immediate neighboring pixels. The textural features provide insights into the temperature distribution patterns within the breast tissue, helping to identify subtle thermal signatures that might be associated with abnormal tissue growth or increased vascularization often seen in cancerous lesions.
[0078] The feature extraction module 210 is configured to extract histogram features by analysing temperature distribution patterns across the preprocessed thermal images. The histogram features extraction process involves creating a histogram of temperature values and deriving key features from it. The feature extraction module 210 calculates the frequency of each temperature value in the image, typically grouping them into bins (e.g., 0.1°C intervals). From this histogram, several features are extracted: the mode (most common temperature), skewness (asymmetry of the distribution), kurtosis (peakedness of the distribution), and percentile values (e.g., 25th, 50th, 75th percentiles).
[0079] It is to be noted that the feature extraction processes described above may be performed in the sequence presented, the order of extraction can be varied or customized based on specific requirements of the breast cancer screening process or characteristics of the thermal imaging data. For instance, the feature extraction module 210 might prioritize spatial feature extraction before statistical analysis in cases where localized anomalies are of particular interest. Alternatively, textural feature extraction using GLCM might be performed first if tissue texture characteristics are deemed most critical for a particular screening scenario. The system 100 allows for flexibility in the feature extraction pipeline, enabling optimization of the process for different clinical needs, imaging conditions, or patient-specific factors.
[0080] The feature extraction module 210 is further configured to generate one or more features vectors based on the extracted one or more temperature-based features, for providing as an input to the one or more deep learning convolution models. The feature extraction module 210 organizes and structures the extracted features into a format that is optimized for machine learning analysis. The feature vectors combine various types of extracted features, including statistical measures (mean, max, min, median, standard deviation), spatial features (hotspot characteristics, symmetry metrics), textural features (GLCM-derived contrast, correlation, entropy), and histogram features (mode, skewness, kurtosis, percentiles).
[0081] The score generating module 212 is configured to generate, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images. The deep learning models process the extracted temperature based features through multiple layers, learning complex patterns and relationships that are indicative of breast abnormalities. The final layer of the neural network outputs a probability score, typically between 0 and 1, representing the likelihood of an abnormality being present. For instance, a score of 0.05 is assigned when there's minimal thermal asymmetry between breasts, no focal hot spots, regular thermal boundaries, and uniform texture patterns. In another instance, a score of 0.75 is assigned when there's significant thermal asymmetry, a pronounced focal hot spot with sharp temperature gradients, irregular thermal boundaries, and distinct texture changes in a localized area. In yet another instance, a score of 0.95 is assigned when there are multiple high-risk features present, such as extreme thermal asymmetry, several focal hot spots with very sharp temperature gradients, highly irregular thermal boundaries, and pronounced texture changes consistent with known malignant patterns.
[0082] The classification module 214 is configured to classify, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score. If the probability score is less than 0.5, then the classification module 214 classifies the breast tissue as normal. If the probability score is 0.5 or higher, then the classification module 214 classifies the breast tissue as abnormal. The threshold of 0.5 is selected based on receiver operating characteristic (ROC) analysis, balancing high sensitivity (greater than or equal to 90 percent) with acceptable specificity. This configuration priorities early detection of abnormalities, which is critical in breast cancer screening applications.
[0083] The display module 216 is configured to display, on the display unit 114 at least one of the thermal images of the breast, the temperature data, classification result associated with the thermal images, and warning message if abnormality is detected, or combination thereof. In some embodiments, the display module 216 employs an intuitive layout with touch-enabled controls for zooming, panning, and adjusting image contrast, while also displaying data privacy indicators and offering secure export options. The comprehensive display ensures that healthcare professionals can quickly interpret results, facilitates patient understanding during consultations, and adapts to various screen sizes and orientations for optimal visibility and usability across different hardware configurations of the edge computing unit 102.
[0084] FIGS. 3A-3B are flow charts illustrating a method of breast cancer screening in accordance with the present disclosure. The process described in FIGS. 3A-3B may be implemented in the system 100 of FIG. 1. For brevity and to avoid redundancy, the individual components of the system 100 will not be re-explained here, as they have been previously detailed in the description of FIG. 1. It is understood that the steps outlined in FIGS. 3A-3B are executed by the relevant modules and components of system 100, utilizing the hardware and software infrastructure described earlier.
[0085] At step 302, the method includes capturing, using at least one thermal camera 108, one or more thermal images of breast of a subject. In one exemplary embodiment, the thermal imaging camera 108 has a resolution of 32 × 24 pixels. For capturing thermal images, the thermal imaging camera 108 is positioned at a predetermined optimal distance and angle from the subject. Further, the camera's field of view is adjusted to ensure both breasts are fully captured within a single frame. In some embodiments, the method includes capturing multiple images in quick succession to account for potential movement artifacts and ensure the best quality image is selected for analysis. It is to be noted that the capture process is standardized, with clear instructions provided to the subject to ensure consistent positioning and environmental conditions. After capturing thermal images, the method includes storing the thermal images in the database 200 or transferring the thermal images to the edge computing unit 102 for further analysis.
[0086] At step 304, the method includes measuring, using at least two temperature sensors 110, 112, single-point temperature at one or more target locations of the breast of the subject. In some embodiments, the temperature sensors 110, 112 are non-contact infrared thermal sensors. While measuring the temperature, the temperature sensors 110, 112 are held at a specific distance from the skin surface, typically 5-10 cm, as recommended by the manufacturer. The subject is asked to remain still during measurements to avoid motion-induced errors.
[0087] At step 306, the method includes selecting, using the verification module 206 of the edge computing unit 102, one or more regions of interest in the thermal images that correspond to the one or more target locations of the breast. At step 308, the method includes averaging or extracting, using verification module 206, temperature values from the one or more regions of interest in the thermal images. At step 310, the method includes comparing, using verification module 206, temperature values from the one or more regions of interest in the thermal images with temperature points in the corresponding target locations of the breast.
[0088] At step 312, the method includes accepting, by the verification module 206, the thermal images for preprocessing if the temperature values from the regions of interest in the thermal images and corresponding temperature points of target locations are matching, otherwise instructing for recapturing thermal images of the breast of the subject.
[0089] At step 314, the method includes preprocessing, using the preprocessing module 208 of the edge computing unit 102, the thermal images by removing noise from the thermal images by applying one or more filters; normalizing pixel intensity values of the thermal images; identifying and isolating breast regions by removing background areas; enhancing contrast of identified and isolated breast regions to improve visibility of subtle temperature differences; resizing enhanced thermal images comprising first pixel values into thermal images comprising second pixel values; and optionally performing color mapping or alignment if multiple views are used.
[0090] At step 316, the method includes extracting, using the feature extraction module 210 of the edge computing unit 102, one or more temperature based features from the thermal images of the breast by extracting mean, maximum, minimum, median, standard deviation of temperatures within the one or more regions of interest; extracting spatial features by performing hotspot analysis and symmetry analysis between left and right breast regions; extracting, using grey-level co-occurrence matrix (GLCM), textural features to assess contrast, correlation, and entropy; and extracting histogram features by analysing temperature distribution patterns across the preprocessed thermal images. At step 318, the method includes generating, using the feature extraction module 210, one or more feature vectors based on the one or more temperature based features.
[0091] At step 320, the method includes calculating, using the score generating module 212 of the edge computing unit 102, the probability score based on preprocessed image and the one or more feature vectors. The deep learning models process the extracted temperature based features through multiple layers, learning complex patterns and relationships that are indicative of breast abnormalities. The final layer of the neural network outputs a probability score, typically between 0 and 1, representing the likelihood of an abnormality being present.
[0092] At step 322, the method includes classifying, using the classification module 214 of the edge computing unit 102, breast tissue associated with the thermal image as normal or abnormal based on the probability score. If the probability score is below 0.5, then the classification module 214 classifies the breast tissue as normal. If the probability score is 0.5 or higher, then the classification module 214 classifies the breast tissue as abnormal.
[0093] At step 324, the method includes displaying, using the display module 216 of the edge computing unit 102, thermal images of the breast, temperature data, classification result associated with the thermal images, and warning message if abnormality is detected.
[0094] In some embodiments, the method further includes recommending treatment based on the classification result. The method interprets the classification result, considering the probability score and other extracted features, to stratify the risk level (e.g., low, moderate, high). Based on this stratification, the method generates tailored treatment recommendations, ranging from routine screening for low-risk cases to urgent specialist referrals or biopsy recommendations for high-risk cases. The recommendations are personalized considering patient-specific factors such as age and family history.
[0095] FIG. 4 is a flow chart illustrating a method of training a deep learning convolution neural network model in accordance with the present disclosure. It is to be noted that, in many cases, the model may be pre-trained on more powerful external computing resources and then installed on the edge computing unit 102 for inference and potential fine-tuning. This approach allows for the development of sophisticated models using large datasets while maintaining the portability and efficiency of the edge computing unit 102 for deployment in various healthcare settings. In some embodiments, the training process depicted in FIG. 4 may be implemented using the system components described in FIG. 1, particularly leveraging the computational capabilities of the edge computing unit 102.
[0096] At step 402, the method includes obtaining a plurality of thermal images of breast. The method performs acquiring a large, diverse dataset of thermal breast images from various sources, encompassing a wide range of breast types, sizes, and pathologies. The dataset includes longitudinal data where possible, relevant metadata, and undergoes quality assurance checks. All images are captured following a standardized protocol to ensure consistency, and the dataset is expanded through controlled augmentation techniques. Care is taken to ensure a balanced representation of normal and abnormal cases, as well as different types of breast abnormalities.
[0097] At step 404, the method includes obtaining temperature data at one or more target locations of corresponding breast tissue. The temperature data may be collected from various sources to create a comprehensive and diverse training dataset for the deep learning models. Sources may include clinical trials, existing medical databases, academic research studies, and data gathered from partner healthcare institutions. The data encompasses a wide range of patient demographics, breast tissue types, and both normal and abnormal cases. The dataset includes precise temperature measurements taken infrared sensors or equivalent devices at standardized locations. Additionally, the dataset may incorporate temperature data from different imaging modalities like high-resolution thermal cameras or dynamic thermography systems to provide a multi-modal perspective. This diverse data collection ensures that the AI models are trained on a broad spectrum of thermal patterns, including subtle variations that might indicate early-stage breast abnormalities.
[0098] At step 406, the method includes verifying temperature values of one or more regions of interest in the thermal images to the temperature data at the corresponding target locations in the breast. If the temperature values are matching, then the thermal images are forwarded to the preprocessing steps.
[0099] At step 408, the method includes preprocessing verified thermal images by removing noise from the thermal images by applying one or more filters; normalizing pixel intensity values of the thermal images; identifying and isolating breast regions by removing background areas; enhancing contrast of identified and isolated breast regions to improve visibility of subtle temperature differences; resizing enhanced thermal images comprising first pixel values into thermal images comprising second pixel values; and optionally performing color mapping or alignment if multiple views are used.
[0100] At step 410, the method includes labeling the preprocessed thermal images as cancerous or noncancerous based on clinical diagnosis. The labeling process involves annotating the calibrated thermal images with diagnosis results from anyone of mammography, ultrasound, biopsy reports, an expert review, or combinations thereof. Regions of interest are annotated for cancerous cases, and each label is assigned a confidence score. The process includes correlation with histopathological findings where available, temporal labeling for longitudinal data, and measures to assess inter-observer variability. The labels are integrated with relevant metadata, including patient history and risk factors, while adhering to strict ethical guidelines and data protection protocols.
[0101] At step 412, the method includes extracting one or more temperature based features from the preprocessed plurality of thermal images by extracting mean, maximum, minimum, median, standard deviation of temperatures within the one or more regions of interest; extracting spatial features by performing hotspot analysis and symmetry analysis between left and right breast regions; extracting, using grey-level co-occurrence matrix (GLCM), textural features to assess contrast, correlation, and entropy; and extracting histogram features by analysing temperature distribution patterns across the preprocessed thermal images.
[0102] At step 414, the method includes training a deep learning model with extracted temperature based features and associated diagnostic labels and preprocessed thermal images. The training step involves feeding the preprocessed dataset into a convolutional neural network (CNN) optimized for thermal image analysis. In some embodiments, the training process utilizes transfer learning, where a pre-trained model on a large dataset is fine-tuned for breast thermal imaging, and data augmentation to enhance model robustness. The model learns to recognize complex patterns and correlations between thermal distributions and breast abnormalities, with a focus on minimizing false positives and negatives. During training, the model is exposed to a diverse range of thermal patterns, including subtle early-stage indicators and more pronounced late-stage thermal signatures. The process employs techniques like batch normalization, dropout, and adaptive learning rates to improve convergence and prevent overfitting. Regular validation checks are performed using a separate dataset to monitor the model's performance and guide hyperparameter tuning. The training also incorporates multi-task learning, simultaneously optimizing for classification accuracy and localization of abnormal regions. In one exemplary embodiment, the deep learning model is VGG-16 (visual geometry group-16). In some embodiments, other machine learning approaches can also be effective. For instance, ensemble methods such as random forests or gradient boosting machines (e.g., XGBoost) can be employed, particularly when working with limited datasets. Support vector machines (SVMs) can also be used. The choice of model depends on factors such as dataset size, computational resources, interpretability requirements, and the specific characteristics of the thermal imaging data. Regardless of the model chosen, the training process involves careful cross-validation, hyperparameter tuning, and performance evaluation using metrics such as accuracy, sensitivity, specificity, and area under the ROC curve (AUC-ROC) to ensure robust and reliable breast cancer screening capabilities.
[0103] FIG. 5 is a block diagram illustrating an edge computing unit 500 in accordance with the present disclosure. The edge computing unit 500 is functionally equivalent to the edge computing unit 102 described in FIG. 1, but with a structural difference. Unlike the edge computing unit 102, where the thermal imaging camera 108, and temperature sensors 110, 112, are communicatively coupled but not physically integrated, device 500 comprises these components within a single unit. Specifically, the edge computing unit 500 comprises a thermal imaging camera 506, a first temperature sensor 508, a second temperature sensor 510, and a display 514, all integrated into the device itself. The edge computing unit 500 comprises a memory 502 storing a set of instructions and a processing unit 504, similar to edge computing unit 102. The functionalities of these components remain the same as in edge computing unit 102, with the thermal imaging camera 506 capturing breast images, sensors 508 and 510 measuring temperature values at one or more target locations, and the processing unit 504 executing the breast cancer screening algorithms. To avoid repetition, the detailed explanations of the screening process and software modules will not be reiterated here, as they are identical to those described for edge computing unit 102. The integrated design of device 500 offers enhanced portability and potentially simplified operation compared to the externally connected components of edge computing unit 102, while maintaining the same breast cancer screening capabilities.
[0104] The system of the present disclosure is advantageous in that the use of thermal imaging provides a non-invasive, radiation-free approach for breast cancer detection. Such that more frequent screenings without risks can be done. The temperature sensors enable precise point measurements, enhancing the accuracy of thermal data. This dual-sensor approach allows for cross-verification of temperature readings, reducing the likelihood of false positives or negatives. The edge computing enables real-time processing of thermal data, reducing latency and allowing for immediate results. The on-site processing also enhances data privacy and security by minimizing data transmission. Further, the verification step enhances the reliability of the thermal data by cross-checking camera readings with sensor measurements. It helps in calibrating the system and identifying any potential errors in thermal imaging. The preprocessing steps improves image quality and standardizes the data, enhancing the accuracy of subsequent analysis. It can reveal subtle thermal patterns that might be obscured in raw images. Due to the preprocessing step, even low resolution thermal images can be used for screening. The deep learning models can identify complex thermal patterns and features that might be missed by human observers or traditional image processing techniques. This improves the sensitivity and specificity of cancer detection. The automated classification reduces the subjectivity in interpretation and can potentially identify early-stage abnormalities that might be missed in visual examination. It also allows for rapid screening of large populations. Overall, this system offers a comprehensive, accurate, and efficient method for breast cancer screening that is safer and potentially more accessible than traditional methods. Its combination of thermal imaging, precise temperature sensing, and AI-driven analysis represents a significant advancement in early cancer detection technology.
[0105] Due to use of affordable thermal imaging camera (e.g., low resolution thermal imaging camera) and temperature sensors compared to high-end medical imaging equipment, the system of the present disclosure is cost effective. This significantly reduces the initial investment and maintenance costs for healthcare providers. Utilizing standard computing hardware (e.g., laptops) as edge computing units reduces the need for specialized, expensive medical-grade computers. This also allows for easy upgrades and replacements using widely available consumer electronics. The real-time processing and analysis reduce the need for multiple patient visits and decrease the overall time and resources required for diagnosis. This increases the efficiency of healthcare delivery and reduces operational costs. The AI-driven analysis minimizes the need for highly trained radiologists to interpret every image, potentially reducing labor costs and addressing the shortage of specialists, especially in underserved areas. The system's portability allows for mobile screening units, reducing the need for fixed, expensive infrastructure. This can significantly lower the cost of providing screening services in rural or remote areas. The system's reliance on AI models and software algorithms means that performance improvements and new features can often be implemented through software updates, reducing the need for hardware replacements. By potentially detecting breast cancer at earlier stages, the system 100 could significantly reduce the costs associated with treating advanced-stage cancers, leading to substantial savings in overall healthcare expenditure. The thermal imaging and sensing components could potentially be used for other medical diagnostic purposes, increasing the return on investment for healthcare providers. Unlike mammography, which requires specialized films or digital plates, the system 100 doesn't rely on consumable materials for each screening, reducing ongoing operational costs. The use of low-power sensors and efficient computing hardware minimizes energy consumption, reducing electricity costs in clinical settings. The system's design allows for easy scaling, from single units in small clinics to multiple units in large hospitals, without a proportional increase in costs.
[0106] The embodiments of the present invention disclosed herein are intended to be illustrative and not limiting. Other embodiments are possible and modifications may be made to the embodiments without departing from the spirit and scope of the invention. As such, these embodiments are only illustrative of the inventive concepts contained herein.
, C , Claims:1. A system (100) for breast cancer screening, comprising:
at least one thermal imaging camera (108, 506) configured for capturing one or more thermal images of a breast of a subject, wherein captured thermal images comprise first pixel values;
at least two temperature sensors (110, 112, 508, 510) configured for measuring temperature values at one or more target locations of the breast of the subject; and
an edge computing unit (102, 500) comprising a memory (104, 502) storing a set of instructions and a processing unit (106, 504),
wherein the thermal imaging camera (108, 506), the at least two temperature sensors (110, 112, 508, 510) are positioned in the edge computing unit (102, 500) or communicatively coupled to the edge computing unit (102, 500),
wherein the processing unit (106, 504) is configured to execute the set of instructions to:
obtain the thermal images of the breast of the subject;
obtain temperature data corresponding to the one or more target locations of the breast of the subject;
verify temperature values of the thermal images based on temperature data corresponding to the one or more locations of the breast of the subject;
preprocess verified thermal images;
extract, using one or more deep learning convolution models deployed in the edge computing unit (102, 500), one or more temperature based features from the preprocessed thermal images;
generate, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images; and
classify, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score.
2. The system (100) as claimed in claim 1, wherein the processing unit (106, 504) to display, on a display unit (114, 512), at least one of the thermal images of the breast, classification result associated with the thermal images, and warning message if abnormality is detected, or combination thereof.
3. The system (100) as claimed in claim 1, wherein the processing unit (106, 504) is configured to verify temperature values of the thermal images by
selecting one or more regions of interest in the thermal images that correspond to the one or more target locations of the breast;
averaging or extracting temperature values from the one or more regions of interest in the thermal images;
comparing temperature values from the one or more regions of interest in the thermal images with corresponding temperature points in the one or more target locations of the breast; and
accepting the thermal images for preprocessing if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are matching.
4. The system (100) as claimed in claim 3, wherein the processing unit (106, 504) is further configured to instruct to capture new thermal images of the breast if the temperature values from the one or more regions of interest in the thermal images and temperature points in the corresponding target locations in the breast, are not matching.
5. The system (100) as claimed in claim 1, wherein the processing unit (106, 504) is configured to preprocess verified thermal images by:
removing noise from the thermal images by applying one or more filters;
normalizing pixel intensity values of the thermal images;
identifying and isolating breast regions by removing background areas;
enhancing contrast of identified and isolated breast regions to improve visibility of subtle temperature differences;
resizing enhanced thermal images comprising first pixel values into thermal images comprising second pixel values; and
optionally performing color mapping or alignment if multiple views are used.
6. The system (100) as claimed in claim 1, wherein the processing unit (106, 504) is configured to extract the one or more temperature-based features from the preprocessed thermal images by:
extracting mean, maximum, minimum, median, standard deviation of temperatures within the one or more regions of interest;
extracting spatial features by performing hotspot analysis and symmetry analysis between left and right breast regions;
extracting, using grey-level co-occurrence matrix (GLCM), textural features to assess contrast, correlation, and entropy; and
extracting histogram features by analysing temperature distribution patterns across the preprocessed thermal images.
7. The system (100) as claimed in claim 6, wherein the processing unit (106, 504) is further configured to generate one or more features vectors based on the extracted one or more temperature-based features, for providing as an input to the one or more deep learning convolution models.
8. The system (100) as claimed in claim 1, wherein the at least two temperature sensors (110, 112, 508, 510) are non-contact infrared thermal sensors.
9. The system (100) as claimed in claim 1, wherein the one or more deep learning convolution models are trained by
obtaining a plurality of thermal images of breast and temperature data at one or more target locations of corresponding breast tissue;
verifying temperature values of one or more regions of interest in the thermal images to the temperature data at the corresponding target locations in the breast;
preprocessing verified thermal images;
labeling the preprocessed thermal images as cancerous or noncancerous;
extracting one or more temperature based features from the preprocessed plurality of thermal images; and
training the one or more deep learning convolution models with extracted features and associated diagnostic labels and preprocessed thermal images
10. A method for breast cancer screening, comprising:
capturing, using at least one thermal camera (108, 506), one or more thermal images of a breast of a subject, wherein captured thermal images comprise first pixel values;
measuring, using at least two temperature sensors (110, 112, 508, 510), temperature values at one or more target locations of the breast of the subject;
verifying, using an edge computing unit (102, 500), temperature values of the thermal images based on temperature data corresponding to the one or more target locations of the breast of the subject;
preprocessing, using the edge computing unit (102, 500), verified thermal images;
extracting, using one or more deep learning convolution models deployed in the edge computing unit (102, 500), one or more temperature based features from the preprocessed thermal images;
generating, using the one or more deep learning convolution models, probability score indicating presence of abnormality associated with breast cancer, based on analysis of extracted temperature based features and preprocessed thermal images; and
classifying, using the one or more deep learning convolution models, breast tissue associated with the thermal images as normal or abnormal based on the probability score.