Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(Section 10 & Rule 13)

AI-POWERED UNIFIED DIAGNOSTIC IMAGE ANALYSIS PLATFORM, SYSTEM AND METHOD THEREOF

APPLICANT NATIONALITY ADDRESS
NUVO AI PRIVATE LIMITED Indian Survey no. 135,139, Bilakhia House, Muktanand Marg, Chala, Vapi, Valsad, Gujarat 396191, India

The following specification particularly describes the invention and the manner in which it is to be performed.

FIELD OF TECHNOLOGY

[0001] The present disclosure generally relates to the field of healthcare systems, and more particularly relates to a method, system and platform for AI-powered unified diagnostic image analysis.

BACKGROUND

[0002] Radiologist use different imaging modalities to capture a medical image and diagnose a condition in an individual using the medical image. X-rays are most commonly used imaging modalities due to their low cost and wide availability.

[0003] Typically, radiologist manually assess medical images of body part or organ to locate abnormality or a condition. Radiologists spend significant time manually scanning for abnormalities, annotating, comparing historical scans, and generating reports. This leads to increase in time to diagnosis of a medical condition, reduced throughput, radiologist fatigue, and human error in diagnosing.

[0004] Also, there is often delay experienced in timely diagnosis and treatment due to rising patient volumes and shortage of trained radiologists. Also, manually interpretation of medical images also causes fragmented access to previous medical records, resulting in inconsistent and error-prone diagnosis of the patient. These issues directly affect patient outcomes especially in trauma centres, Intensive Care Units, emergency departments and rural healthcare networks.

SUMMARY

[0005] An artificial intelligence (AI) powered unified diagnostic image analysis platform, system and method is disclosed. In one aspect, a method includes obtaining a medical image associated with an individual from a medical image database, pre-processing, via the unified diagnostic image analysis platform, the medical image associated with the individual, and performing, via the unified diagnostic image analysis platform, a plurality of operations by applying the pre-processed medical image to a plurality of trained artificial intelligence models. The plurality of operations includes detecting one or more abnormalities associated with the body part by applying the pre-processed medical image to an ensemble of trained abnormality detection models, classifying a condition associated with the individual based on the detected abnormalities by applying the pre-processed medical image to a trained condition classification model, and generating at least one explainability output based on at least one of the detected abnormalities and the classified condition associated with the individual.

[0006] The method may include substantially simultaneously performing, via the unified diagnostic image analysis platform, the plurality of operations by applying the pre-processed medical image to the trained artificial intelligence models.

[0007] The method may include determining whether the pre-processed medical image associated with the individual is an inverted medical image, and generating a standard medical image using the pre-processed medical image if the medical image is the inverted medical image.

[0008] The method may include detecting presence of the body part in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model.

[0009] The method may include performing segmentation of the organs in the pre-processed medical image by applying the pre-processed medical image to a trained organ segmentation model.

[00010] The method may include generating a bone-suppressed medical image of the individual by applying the pre-processed medical image to a trained bone suppression model.

[00011] The method may include generating a heatmap representing heatmap activation patterns corresponding to the detected abnormalities associated with the body part.

[00012] In another aspect, an apparatus includes a processing unit, and a memory unit communicatively coupled to the processing unit. The memory unit includes an artificial intelligence (AI) powered unified diagnostic image analysis platform. The unified diagnostic image analysis platform is configured to obtain at least one medical image associated with an individual from a medical image database, pre-process the medical image associated with the individual, and perform a plurality of operations by applying the pre-processed medical image to a plurality of trained artificial intelligence models. The unified diagnostic image analysis platform performs an operation by detecting one or more abnormalities associated with the body part by applying the pre-processed medical image to an ensemble of trained abnormality detection models. The unified diagnostic image analysis platform performs an operation by classifying a condition associated with the individual based on the detected abnormalities by applying the pre-processed medical image to a trained condition classification model. The unified diagnostic image analysis platform performs an operation by generating at least one explainability output based on at least one of the detected abnormalities and the classified condition associated with the individual.

[00013] The unified diagnostic image analysis platform may be configured to substantially simultaneously perform the plurality of operations by applying the pre-processed medical image to the trained artificial intelligence models.

[00014] The unified diagnostic image analysis platform may perform an operation by determining whether the pre-processed medical image associated with the individual is an inverted medical image, and generate a standard medical image using the pre-processed medical image if the medical image is the inverted medical image.

[00015] The unified diagnostic image analysis platform may perform an operation by detecting presence of the body part in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model.

[00016] The unified diagnostic image analysis platform may perform an operation by performing segmentation of organs in the pre-processed medical image by applying the pre-processed medical image to a trained organ segmentation model.

[00017] The unified diagnostic image analysis platform may perform an operation by generating a bone-suppressed image of the individual by applying the pre-processed medical image to a trained bone suppression model.

[00018] The unified diagnostic image analysis platform may perform an operation by generating a heatmap representing heatmap activation patterns corresponding to the detected abnormalities associated with the body part.
[00019] In yet another aspect, a non-transitory computer-readable storage medium, having machine-readable instructions stored therein, that when executed by a processing unit, cause the processing unit to perform method described above.

BRIEF DESCRIPTION OF THE DRAWINGS

[00020] The drawings described herein are for illustrative purposes and are not intended to limit the scope of the present subject matter in any way:

[00021] FIG. 1 illustrates a block diagram of a unified diagnostic image analysis platform using trained artificial intelligence (AI) models, according to one embodiment;

[00022] FIG. 2 illustrates various components of an image pre-processing engine of FIG. 1, according to one embodiment;

[00023] FIG. 3 is a block diagram depicting components of an AI analysis engine of FIG. 1, according to one embodiment;

[00024] FIG. 4 is a block diagram depicting components of an image post-processing engine of FIG. 1, according to one embodiment;

[00025] FIG. 5 is a process flowchart depicting an exemplary method of analysing diagnostic images using the unified diagnostic image analysis platform, according to one embodiment;

[00026] FIG. 6 is a block diagram of an exemplary cloud computing environment for performing analysis of diagnostic images using trained AI models, according to one embodiment;

[00027] FIG. 7 is a block diagram of an exemplary server-based computing system for performing analysis of diagnostic images using trained AI models, according to another embodiment;

[00028] FIG. 8 is a block diagram of computing system implementing a unified diagnostic image analysis platform for performing analysis of diagnostic images using trained AI models, according to yet another embodiment;
[00029] FIG. 9 is a flow diagram depicting analysis of a chest X-ray image by the unified diagnostic image analysis platform, according to an exemplary implementation;

[00030] FIG. 10A is a user interface view depicting a medical image with abnormalities detected by the unified diagnostic image analysis platform, according to one embodiment; and

[00031] FIG. 10B is a user interface view depicting a heatmap generated by the unified diagnostic image analysis platform, according to one embodiment.

DETAILED DESCRIPTION

[00032] Examples described herein may provide an AI-powered unified diagnostic image analysis platform, system and method thereof. Various embodiments are described with reference to the drawings, wherein like reference numerals are used to refer the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for the purpose of explanation, numerous specific details are set forth in order to provide thorough understanding of one or more embodiments. It may be evident that such embodiments may be practiced without these specific details.

[00033] The terms ‘medical images’ and ‘diagnostic images’ means the same and are used interchangeably throughout the document.

[00034] Also, the terms ‘body part’, ‘organ’, and ‘anatomical structure’ means the same and are used interchangeably throughout the document.

[00035] FIG. 1 illustrates a block diagram of a unified diagnostic image analysis platform 100 for performing analysis of diagnostic images using trained artificial intelligence models, according to one embodiment. The unified diagnostic image analysis platform 100 includes a medical image database 102, an image pre-processing engine 104, an AI model database 106, an AI analysis engine 108, an image post-processing engine 110, a report database 112, a patient database 114, and a visualization module 116.

[00036] The medical image database 102 may be configured to store medical images of individuals (e.g., patient) for diagnostic image analysis. The medical images may be in DICOM or other compatible formats. The medical images may be acquired using one or more imaging modalities in a healthcare facility. The imaging modalities may include X-ray machines, MRI, Ultrasound, Computer Tomography, etc.

[00037] The image pre-processing engine 104 may be configured to pre-process the medical images including adjusting contrast, size, inverting the medical image, detecting presence of anatomical structures, etc. The AI model database 106 may be configured to store a plurality of specially trained artificial intelligence models for performing a plurality of operations on the medical images along the diagnostic image analysis pipeline. For example, the trained artificial intelligence models may include an organ detection model, an abnormality detection model, a condition classification model, an organ segmentation model, a bone suppression model, and the like. Each trained AI model may be executed by applying features of the medical image to perform an operation including but not limited to image classification, organ detection, abnormality detection, condition classification, organ segmentation, bone suppression, etc. In some embodiments, the trained AI models may be ensemble of trained AI models to perform the diagnostic image analysis including but not limited to detecting a plurality of abnormalities in anatomical structures of an individual.

[00038] The AI analysis engine 108 is configured to perform one or more operations related to diagnostic image analysis using the one or more trained AI models stored in the AI model database 106. In some embodiments, the AI analysis engine 108 is configured to substantially simultaneously perform a plurality of operations on medical images using the trained AI models. For example, if hundreds of medical images from imaging modalities in different healthcare facility are uploaded to the medical image database 102 in an hour, the AI analysis engine 108 is configured to substantially simultaneously perform analysis of the medical images pertaining to different individuals (e.g., having unique patient identifiers) in real time using the trained AI models. In these embodiments, the AI analysis engine 108 may analyse a plurality of medical images of the different individuals in a single instance. The output of the AI analysis engine 108 may be labelled using a unique identifier associated with the individuals being diagnosed. Advantageously, time to analyse medical images by a radiologist or individual computer system (e.g., desktop computer) is substantially reduced.

[00039] In one embodiment, the AI analysis engine 108 is configured to simultaneously detect multiple abnormalities associated with the individual by applying the pre-processed medical image(s) to an ensemble of trained abnormality detection models. For example, if the medical image is associated with thorax area of the individual, the AI engine 108 is capable of detecting abnormalities such as opacity, lung nodule, rib fracture, pleural effusion, etc. using an ensemble of trained abnormality detection models. In another embodiment, the AI analysis engine 108 is configured to classify a condition associated with the individual(s) based on the determined abnormalities associated with the body part by applying the pre-processed medical image(s) to the trained condition classification model. For example, if the medical image is associated with the thorax area of the individual, the AI analysis engine 108 may determine tuberculosis, lung cancer, cardiomegaly, etc. by applying the pre-processed medical image(s) to the trained condition classification model.

[00040] In yet another embodiment, the AI analysis engine 108 is configured to perform segmentation of the organs in the pre-processed medical image(s) by applying the pre-processed medical image(s) to the trained organ segmentation model. For example, the AI analysis engine 108 may generate organ segmentation masks such as heart mask or lung masks which designate anatomical region associated with specific organ or body part. In further another embodiment, the AI analysis engine 108 is configured to generate bone suppressed images associated with individual(s) by applying the pre-processed medical image(s) to the trained bone suppression model. For example, the bone suppressed images enable user to clearly view the abnormalities associated with anatomical structures. In some embodiments, the AI engine 108 is configured to generate bone suppressed images using the trained bone suppression model in such a manner that bony structures in the medical images are attenuated while preserving pathological structures and/or abnormalities such as nodules overlapping the bony structures.

[00041] In one embodiment, the image post-processing engine 110 is configured to post-process the output of the image preprocessing engine 104, and the AI analysis engine 108, including but not limited to organ detection, abnormality detection, condition classification, organ segmentation, to generate an explainable output. The explainable output may provide transparent and understandable output of the AI analysis engine 108 such as highlighting specific regions associated with the detected abnormalities and/or detected conditions via heatmap activation patterns, computing a confidence score associated with the detected abnormalities, computing cardiothoracic ratio supporting detection of abnormality such as cardiomegaly, etc. In one embodiment, the image post-processing engine 110 may generate a heatmap associated with the organ of an individual using feature maps corresponding to the detected abnormalities, classified condition, and organ segmentation masks. The heatmap associated with the anatomical structures may include explainability indicators such as heatmap activation patterns corresponding to the detected abnormalities and the detected conditions. The image post-processing engine 110 may restrict heatmap activation patterns to relevant anatomical region in the heatmap using the organ segmentation masks. The image post-processing engine 110 may blend the heatmap with the original medical image (e.g., 50% opacity) for overlay visualization.

[00042] The image post-processing engine 110 may compute cardiothoracic ratio (CTR) of the individual based on the detected organ and the organ segmentation masks (e.g., heart mask). In another embodiment, the image post-processing engine 110 may generate abnormality score based on the detected abnormalities associated with body part.

[00043] In further another embodiment, the image post-processing engine 110 may generate clinical grade structured report indicating clinical findings, confidence scores, clinical summaries, severity scores, visual evidence highlighting region of interests, etc. The image post-processing engine 110 is configured to generate interactive reports with directly linked annotations, region to sentence linkage and editable radiology notes. The report database 112 is configured to store reports associated with individuals based on the diagnostic image analysis. For example, the report database 112 may store standard reports, detailed clinical reports, and full screening reports.

[00044] The visualisation module 116 is configured to provide access to multi-format reports to clinicians, radiologist, patients, doctors, etc using the reports stored in the report database 112 and information from the patient database 114. For example, a radiologist may enter a patient identifier in a graphical user interface of the unified diagnostic image analysis platform to access reports of the patient associated with the patient identifier. The visualisation module 116 may verify the patient identifier with patient identification data stored in the patient database 114. Once the patient identifier is verified, the visualisation module 116 may fetch corresponding reports from the report database 112 and render the reports on the graphical user interface. The visualisation module 116 may render information in the reports based on a role and privileges. For example, the visualisation module 116 may display a full detailed screening report to the radiologist and display standard report to a patient. The visualisation module 116 may enable a radiologist to simultaneously view original, bone-supressed, heatmap-overlay, CTR-related views with synchronized zoom, pan and interaction controls. The parallel evaluation enables deeper clinical insight without switching tools or systems, thereby reducing cognitive load on radiologists and interpretation time.

[00045] The visualization module 116 may also provide historical patient comparison and dashboard analytics. In this manner, the visualisation module 116 provides evidence linked intelligent reporting integrated with live annotations and region to sentence linkage. The visualisation module 116 may also enable a radiologist to edit radiology notes. The visualisation module 116 may also provide diagnostic recommendations based on patterns learnt from a radiologist including preferred annotation types, unique reporting phrasing, region of interest commonly inspected, personal threshold for criticality, etc.

[00046] The unified diagnostic image analysis platform 100 provides end-to-end analysis of medical images with explainability outputs by use of trained artificial intelligence models. The unified diagnostic image analysis platform 100 can be implemented on a cloud computing system in cases where healthcare facility is enabled via the Internet. In such case, the unified diagnostic image analysis platform 100 can be accessed using laptop, tablet or smartphone. The unified diagnostic image analysis platform 100 can be implemented on an offline inference box where the healthcare facility is not enabled via the Internet.

[00047] FIG. 2 illustrates various components of the image pre-processing engine 104, according to one embodiment. The image pre-processing engine 104 includes a pre-processing module 202, an image inversion module 204, and an organ detection module 206.

[00048] The pre-processing module 202 is configured for pre-processing a medical image (e.g., DICOM image) of an individual. The medical image is captured by an imaging modality (e.g., X-ray machine) to diagnose an organ or body part of the individual (e.g., patient). The medical image may be captured and stored in the medical image database 102 in DICOM, PNG or other standard formats. The pre-processing module 202 may crop a region of interest in the medical image. The pre-processing module 202 may enhance contrast of the medical image. In an exemplary implementation, the pre-processing module 202 may adjust contrast of the medical image using Contrast Limited Adaptive Histogram Equalization Technique (CLAHE) technique. Advantageously, local contrast in under-exposed, noisy or low-quality medical images is improved, thereby enabling better visualisation of faint abnormalities. The pre-processing module 202 may resize the cropped medical image to a pre-defined size (e.g., 224 x 224) while ensuring that aspect ratio is maintained.

[00049] The image inversion module 204 is configured for determining whether the medical image is an inverted medical image. In one exemplary implementation, the image inversion module 204 determines whether the medical image is an inverted medical image or standard medical image using the metadata of the medical image. The presence of photometric interpretation in the metadata of the medical image indicates that the medical image is a standard medical image. The image inversion module 204 is configured to generate a standard medical image based on the pre-processed medical image if the medical image is determined as the inverted medical image. In case of X-ray image, the image inversion module 204 may compute a negative pixel value (negative pixel value = maximum pixel value – original pixel value) corresponding to each of the pixels in the inverted X-ray image. The image inversion module 204 generates a standard X-ray image associated with the individual based on the negative pixel value corresponding to said each pixel in the inverted X-ray image.

[00050] The organ detection module 206 is configured for detecting presence of the organ in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model. The trained organ detection model detects an organ in the medical image by assigning a bounding box along with a confidence score. In case, multiple confidence scores are assigned to the organ in the medical image, the organ detection module 206 selects the bounding box with a label (e.g., organ name or ID) with highest confidence score. In case of diagnostic image analysis of chest, the organ detection module 206 is configured to detect lungs in a chest X-ray using a trained lung detection model. The trained lung detection model may be a real-time detection transformer (RT-DETR) model. In an example, the RT-DETR model is trained to explicitly locate boundaries of lungs in the chest X-ray image using a training dataset consisting of 900 chest X-ray images annotated with bounding boxes around lung fields. Based on the training dataset, the RT-DETR model learns global context and spatial dependency of lung anatomy. During inference, the RT-DETR model identifies lung boundaries in the chest X-ray image by recognizing characteristic features (e.g., rib cage boundaries, diaphragm, lung apices). The trained RT-DETR model assigns bounding box coordinates in the chest X-ray image encompassing the boundaries of the lungs and computes a confidence score to the bounding box.
[00051] FIG. 3 is a block diagram depicting components of the AI analysis engine 108 of FIG. 1, according to one embodiment. The AI analysis engine 108 includes an abnormality detection module 302, a condition classification module 304, an organ segmentation module 306, and a bone suppression model.

[00052] The abnormality detection module 302 is configured to detect presence of abnormality in the detected organ of the individual by applying the pre-processed medical image of the organ to a trained abnormality detection model 301. In an embodiment, the trained abnormality detection model 301 may be an ensemble of trained abnormality detection models trained for detecting abnormalities using features of the pre-processed medical image. The ensemble of trained abnormality detection models may be multi-head real-time detection transformer model configured to substantially simultaneously detect multiple abnormalities in an anatomical structure (e.g., thorax area). For example, the multi-head real-time detection transformer model can detect lung nodule, opacity, rib fracture, cardiomegaly, pneumothorax, etc. from a chest X-ray image. One or more heads of the multi-head real-time detection transformer model may be capable of determining multiple abnormalities. It can be noted that the abnormality detection module 302 may run various trained abnormality detection models to detect abnormalities in anatomical structures of individuals.

[00053] The condition classification module 304 is configured to classify a condition associated with the individual by applying the pre-processing medical image to a trained condition classification model 303. The trained condition classification model 303 may be a deep learning model trained for classifying a condition based on the abnormality detected in the organ. For example, the trained condition classification model 303 may be a trained TB detection model, a trained lung cancer detection model, a trained breast cancer detection model, and the like. In one embodiment, the condition classification module 304 may compute probability of an individual suffering from tuberculosis using a trained TB detection model based on pulmonary consolidation detected in the lung region. In another embodiment, the condition classification module 304 may compute probability of an individual suffering from lung cancer using a trained lung cancer detection model based on a lung nodule detected in the lung region. The condition classification module 304 may detect different kinds of conditions or diseases in individuals using one or more artificial intelligence models trained to classify a specific condition in individuals. In some embodiments, the one or more artificial intelligence models are trained to classify specific conditions using X-ray images. In other embodiments, the one or more artificial intelligence models are trained to classify specific conditions using other types of diagnostic images such as CT images, MRI images, Ultrasound images.

[00054] The organ segmentation module 306 is configured to segment organs in the pre-processed medical image using a trained organ segmentation model 305 stored in the AI models database 106. The trained organ segmentation model 305 may be a deep learning model trained to perform pixel-wise semantic segmentation of the pre-processed medical image to identify organs in the pre-processed image. For example, the organ segmentation module 306 may perform pixel-wise semantic segmentation of fourteen or more anatomical structures in a chest X-ray image including lungs, heart, mediastinum, spine and associated thoracic regions using the trained organ segmentation model 305.

[00055] The organ segmentation module 306 may generate binary organ segmentation masks from the pre-processed medical image of the individual. In an exemplary implementation, the organ segmentation module 306 generates binary organ segmentation masks by applying the pre-processed medical image to a trained Pyramid Scene Parsing Network (PSPNet) model. The trained PSPNet model consists of an encoder, pyramid pooling module, and a decoder. The encoder may be ResNet convolutional network which processes the pre-processed medical image and hierarchical spatial feature maps at progressively lower resolutions. In the encoder, deep layers capture semantic features (e.g., regarding structure) and shallow layers retain spatial details (e.g., boundaries). The pyramid pooling module applies four parallel pooling operations at different scales (1x1, 2x2, 3x3 and 6x6) to the hierarchical spatial feature maps. Each pooling operation captures context at a different spatial granularity. For example, the 1x1 pooling produces a single global average encoding overall distribution of structures across the entire medical image. The 2x2 and 3x3 poolings capture regional context (e.g., relative positioning of heart and lungs). The 6x6 pooling captures finer local context. The outputs of all four pooling operations are upsampled to the encoder feature map size. The upsampled pooling operations are concatenated. This helps to achieve simultaneous awareness of fine local anatomical structure and global anatomical layout in a single representation, esp. when position of one anatomical structure (e.g., mediastinum) is highly informative about expected position and extent of adjacent anatomical structures (e.g., lungs, hilum). The decoder produces per-pixel predictions at 512x512 resolution using fourteen independent sigmoid activations (one per anatomical structure). The sigmoid function is capable of differentiating pixels which legitimately belong to multiple anatomical structures (spine overlapping with mediastinum). The sigmoid output produces continuous probability values in [0,1] for each pixel and each class. The probability values are binarized using a threshold of 0.5. That is, the pixels with probability values greater than or equal to 0.5 are assigned to respective anatomical structures. The threshold of 0.5 is a natural decision boundary for a sigmoid trained with binary cross-entropy corresponding to equal weighting of false positives and false negatives. However, per-class thresholds can be tuned if sensitivity/specificity trade-offs differ by anatomical structure.

[00056] Typically, Blood vessels, bronchial walls, and pathological opacities such as consolidation and masses may appear as high-attenuation regions within the lung field on a chest X-ray. The high-attenuation regions may be classified as not belonging to lung, thereby creating enclosed void regions inside solid lung mask. These holes are clinically incorrect and cause errors during downstream processing. There may be also under-segmentation at pleural margins particularly where lung parenchyma transitions to pleural space or chest wall. This may cause the mask to be marginally smaller than true lung extent.

[00057] The organ segmentation module 306 computes bounding region of interest from the organ segmentation mask outputted by the trained PSPNet model. The bounding region of interest confines subsequent operations to specific region, thereby avoiding spurious interactions with background regions. The organ segmentation module 306 performs morphological dilation on consolidated binary organ segmentation mask. For example, morphological dilation is performed using 5x5 square structuring element by replacing each pixel with maximum value in 25-pixel neighbourhood. Consequently, the organ segmentation mask is extended outwards by 2 pixels in all directions (horizontal, vertical, and diagonal). The structural expansion smooths boundary of merged components, recovers slightly under-segmented pleural margins and ensures that the final organ segmentation mask strictly encompasses true anatomical organ boundary.

[00058] The organ segmentation module 306 generates a solid, contiguous organ segmentation mask by filling enclosed void regions (holes) within the organ boundary. The organ segmentation module 306 upscales the binary organ segmentation mask to input medical image resolution using nearest-neighbour interpolation.

[00059] In one example, the binary mask may be used for measurement of cardiac width from heart segmentation and computing cardiothoracic ratio (CTR). In another embodiment, the binary mask may be used for determining organ on which bounding box is located. In yet another example, the binary masks may be combined to spatially constrain and refine heatmaps.

[00060] The bone suppression module 308 is configured to generate bone suppressed images from the pre-processed medical image using a trained bone suppression model. In an embodiment, the trained bone suppression model is a deep learning model (e.g., auto-encoder model) trained to generate bone suppressed images from chest X-ray images by attenuating bony structures in the chest X-ray images and preserving pathological structures including nodules (e.g., lung nodules) overlapping the bony structures in the chest X-rays. Advantageously, the bone suppressed images with preserved pathological structures enables radiologist to clearly identify abnormalities in anatomical structures.

[00061] FIG. 4 is a block diagram depicting components of the image post-processing engine 108 of FIG. 1, according to one embodiment. The image post-processing engine 108 includes a score computation module 402, an explainability module 404, a report generation module 406, and a recommendation module 408.

[00062] The score computation module 402 is configured to compute an abnormality score based on the detected abnormality in the organ of the individual. The abnormality score may indicate severity associated with the abnormality and risk to the individual. In one embodiment, the score computation module 402 computes a classification score (Sclass) indicating probability that the detected abnormality belonging to a specific class. In another embodiment, the score computation module 402 computes an intersection over union (IoU) score (SIoU) based on overlap between different bounding boxes corresponding to same abnormality. For example, IoU score is computed as a ratio of area of intersection divided by area of union. The score ranges from 0 to 1, where higher IoU score indicates better localization accuracy. The score computation module 402 determines a spatial weighing mask (Wanatomical) based on anatomical landmarks associated with specific abnormalities. The Wanatomical gives slightly higher weights for specific areas of an organ (e.g., lungs) related to the detected anomaly. For example, if a pleural effusion generally occurs in the bottom of lungs, Wanatomical provides higher weights to the lower regions of the lungs. Accordingly, the score computation module 402 computes an abnormality score based on the classification score (Sclass), IoU score (SIoU) and spatial weighting mask (Sanatomical). In this manner, the abnormality score is recalibrated using Sanatomical, thereby providing near-accurate probability of an abnormality detected in an anatomical structure. In another embodiment, the score computation module 402 is configured to compute a score associated with the classified condition of the patient. The score associated with the classified condition may indicate probability associated with classified condition (e.g., TB).

[00063] The explainability module 404 is configured to generate a multi-dimensional explainability output explaining diagnostic image analysis of the AI analysis engine 104. In one embodiment, the explainability module 404 is configured to generate explainability overlays including but not limited to heatmap overlays, segment overlays, CTR measurements, and so on. In an example, heatmap overlays may include saliency maps. In an embodiment, the explainability module 404 is configured to generate a heatmap associated with an anatomical structure (e.g., lungs) using feature maps corresponding to the detected abnormality (e.g., nodule) in the anatomical structure. The heatmap associated with the anatomical structure may include heatmap activation patterns corresponding to the detected abnormality in the anatomical structure. In an example, the explainability module 404 generates a union heatmap of the enhanced feature maps of the detected abnormality in the anatomical structure. The explainability module 404 resizes the heatmap to match original medical image dimensions (e.g., 1024x1024).

[00064] The explainability module 404 is configured to enhance the heatmap associated with the anatomical structure of the individual by restricting the heatmap activation patterns to region corresponding to an anatomical mask (e.g., lung mask). In an example, the explainability module 404 applies the anatomical mask to the heatmap, thereby restricting heatmap activation patterns to relevant anatomical region (e.g., lung region). The explainability module 404 dynamically computes intensity threshold based on non-zero pixel statistics. Furthermore, the explainability module 404 zero-out low-intensity values within the relevant anatomical region. The explainability module 404 blends the enhanced heatmap with original medical image (e.g., 50% opacity) for overlay visualization. Additionally, the explainability module 404 is configured to generate CLAHE enhanced medical images. The explainability module 404 is configured to compute cardiothoracic ratio using chest width from bounding box coordinates of the lungs and heart width from an anatomical mask (e.g., heart mask). For example, the explainability module 404 provides explainability via cardiothoracic ratio (>0.5) supporting the detected abnormality such as cardiomegaly.

[00065] The report generation module 406 is configured to generate evidence-backed clinical grade structured reports using a trained reporting model (e.g., trained LLM model). In an embodiment, the report generation module 406 enables generation of multi-template, multi-format reports including but not limited to standard reports, detailed clinical reports, and full screen reports. The report generation module 406 enables automated downloading of clinical grade structured reports. In one embodiment, the reports may include findings from diagnosis of organ of the individual, confidence scores such as abnormality score and condition score, clinical summary, visual evidence, etc. In another embodiment, the reports may be interactive reports with directly linked live annotations, region to sentence linkage, editable radiology notes, and patient friendly. In yet another embodiment, the reports may indicate severity levels, highlight region of interests in the medical image, and a quick clinical summary.

[00066] The recommendation module 408 is configured to generate personalized recommendations for a radiologist based on specific radiologist patterns using a trained recommendation model (e.g., trained LLM model). In an example, the recommendation module 408 may learn radiologist patterns in real-time and recommend annotation type, reporting phrases, commonly inspected region of interest, personal threshold for criticality.

[00067] FIG. 5 is a process flowchart depicting an exemplary method of analysing diagnostic images using a unified diagnostic image analysis platform 100, according to one embodiment. At act 502, a medical image associated with the individual is obtained from a medical image database (e.g., the medical image database 102). The medical image may correspond to a body part/organ of an individual (e.g., chest/thorax of a patient). The medical image (e.g., chest X-ray) is captured using an imaging modality (e.g., X-ray). The medical image may be in DICOM or other compatible formats.

[00068] At act 504, the medical image associated with the individual is pre-processed. For example, the chest X-ray may be cropped and enhanced using CLAHE technique. Also, the medical image may be transformed into a standard medical image in case the medical image is an inverted medical image. The presence of organ or body part (e.g., lungs) in the medical image (e.g., chest X-ray) to be analysed may be determined. One or more of the pre-processing steps may be performed using trained artificial intelligence models (e.g., trained organ detection model).

[00069] At act 506, an abnormality associated with the body part is determined by applying the pre-processed medical image to a trained abnormality detection model (e.g., the abnormality detection model 301). At act 508, a condition associated with the individual is classified by applying the pre-processed medical image to a trained classification model (e.g., the trained condition classification model 303).

[00070] At act 510, an explainability output is generated based on the determined abnormality in the body part and/or the classified condition associated with the individual. The explainability output may include heatmaps, bone suppressed views, bounding boxes, segmented maps and the like. The explainability output may also include historical patient comparison and progress over time. The explainability output enables radiologists to understand rationale behind outcome of the AI based analysis of the medical image.

[00071] At act 512, the explainability output associated with the body part of the individual is displayed on a graphical user interface. Additionally, a clinical grade structured report associated with analysis of the medical image is displayed on the graphical user interface. The report may include clinical findings, confidence scores on abnormality and conditions, clinical summary, visual evidence. The clinical summary may include quick clinical summary, and severity levels. The visual evidence may highlight region of interest in the medical image. The report may also include directly linked annotations, region to sentence linkage and editable radiology notes.

[00072] Although method acts 506 to 510 are illustrated in sequential manner, it can be noted that the unified diagnostic image analysis platform 100 may perform the acts 506 to 508 substantially simultaneously, thereby speeding up the process of analysing diagnostic images.

[00073] FIG. 6 is a block diagram of an exemplary cloud computing environment 600 for AI-powered unified diagnostic image analysis, according to one embodiment. The cloud computing environment 600 may include a cloud computing system 602 and a plurality of devices 612A-N connected to the cloud computing system 602 via a network 610.

[00074] The cloud computing system 602 may include a diagnostic image analysis platform 100, a cloud computing hardware 606, and a cloud communication interface 608. The cloud communication interface 608 may enable communication between the diagnostic image analysis platform 100 and the devices 612A-N via the network 610. The cloud computing hardware 606 may include one or more servers on which an operating system is installed and including one or more processors, one or more storage devices for storing data, and other peripherals required for providing cloud computing functionality such as AI-powered unified diagnostic image analysis and reporting. The diagnostic image analysis platform 100 may be a platform which is capable of delivering functionalities such as data storage (e.g., medical image database 102, AI model database 106, report database 112, and patient database 114), real-time data analysis (e.g., inversion of medical image, organ detection, detecting abnormality in organ, classifying condition of organ, organ segmentation), data visualization (e.g., explainability overlays, interactive clinical grade structured reports, etc.), data communication (e.g., receive a request from the devices 612A-N to perform analysis of a medical image and rendering a report for respective patient based on analysis in real-time) using cloud computing hardware 606 via application programming interfaces (APIs), and capable of delivering the aforementioned cloud service.

[00075] As shown in FIG. 1, the diagnostic image analysis platform 100 includes an image pre-processing engine 104, an AI analysis engine 108, an image post-processing engine 110, and visualization module 116. The unified diagnostic image analysis platform 100 is deployed in the form of machine-readable instructions, that when executed by the cloud computing hardware 606, cause the diagnostic image analysis platform 100 to perform analysis of diagnostic images using multiple trained artificial intelligence models, generate explainability output based on analysis of the medical images, and render reports including explainability output on a respective graphical user interface of the devices 612A-N.

[00076] FIG. 7 is a block diagram of an exemplary server-based computing system 700 for performing analysis of diagnostic images using trained AI models, according to another embodiment. In some embodiments, the server-based computing system 100 is implemented on an offline inference box. The server-based computing system 700 includes a server 702, and a plurality of devices 706A-N. The plurality of devices 706A-N are connected to the server 702 via a network 704. The server 702 includes the unified diagnostic image analysis platform 100. The server 702 may include a processor, a memory, and a storage unit. The diagnostic image analysis platform 100 may be stored on a memory in the form of machine-readable instructions executable by the processor. The server 702 may include a communication interface for enabling communication with the devices 706A-N via the network 704.

[00077] As shown in FIG. 1, the unified diagnostic image analysis platform 100 includes an image pre-processing engine 104, an AI analysis engine 108, an image post-processing engine 110, and visualization module 116. The diagnostic image analysis platform 100 is stored in the form of machine-readable instructions, that when executed by the server 702, cause the diagnostic image analysis platform 100 to perform analysis diagnostic images using multiple trained artificial intelligence models, generate explainability output based on analysis of the medical images, and render reports including explainability output on a respective graphical user interface of the devices 706A-N.

[00078] Method acts performed by the server 702 to achieve the above-mentioned functionality are described in greater detail in FIG. 5. The devices 706A-N include graphical user interfaces providing a request to perform analysis of a medical image of a patient and visualizing interactive report on analysis of the medical image. Each of the devices 706A-N may be provided with a communication interface for interacting with the server 702. Users (doctors, hospital staff) of the devices 706A-N may access the server 702 via the graphical user interfaces. For example, the users may send a request to the server 702 to perform analysis of a medical image, generation of a report, downloading of report and so on. The graphical user interfaces may be specifically designed for accessing the unified diagnostic image analysis platform in the server 702.

[00079] FIG. 8 is a block diagram of computing system 800 implementing a unified diagnostic image analysis platform 100 for performing analysis of diagnostic images using trained AI models, according to yet another embodiment. The computing system 800 may be a personal computer, a laptop, smartphone and the like. The computing system 800 includes a processing unit 802, a memory unit 804, a bus 808, an input unit 810, and a display unit 812.

[00080] The processing unit 802, as used herein, may be any type of computational circuit such as but not limited to a microprocessor, reduced instruction set computing microprocessor, very long instruction word processor, explicitly parallel instruction computing microprocessor, graphics processor, digital signal processor, or any type of processing circuit. The processing unit 802 may also include embedded controllers, such as generic or programmable logic devices or arrays, application specific integrated circuits, single-chip computers and the like.

[00081] The memory unit 804 may be non-transitory volatile memory and non-volatile memory. The memory unit 804 may be coupled for communication with the processing unit 802, such as being a computer readable storage medium. The processing unit 802 may execute instructions and /or code stored in the memory unit 804. A variety of machine-readable instructions may be stored in and accessed from the memory unit 804. The memory unit 804 may include any suitable elements for storing data and machine-readable instructions such as read only memory, random access only memory, erasable programmable read only memory, electrically erasable programmable read only memory, a hard drive, a removable media drive for handling compact disks, digital video disks, diskettes, magnetic tape cartridges, memory cards, and the like. In the present embodiment, the memory unit 804 includes a unified diagnostic image analysis platform 100 of FIG. 1 stored in the form of machine-readable instructions on any of the above-mentioned storage media and may be in communication to and executed by the processing unit 802. When executed by the processing unit 802, the unified diagnostic image analysis platform 100 cause the processing unit 802 to perform analysis diagnostic images using multiple trained artificial intelligence models, generate explainability output based on analysis of the medical images, and render reports including explainability output on a display unit 810.

[00082] The input unit 810 may include input devices such as a keypad, a touch sensitive display capable of receiving request for performing analysis of diagnostic images using trained AI models. The display unit 812 may be a graphical user interface displaying an interactive report of analysis of diagnostic images. The bus 808 acts as an interconnect between the processing unit 802, the memory unit 804, the input unit 810, and the display unit 812.

[00083] Those of ordinary skilled in the art will appreciate that the hardware depicted in FIG. 8 may vary for particular implementations. For example, other peripheral devices such as an optical disk drive and the like, Local Area Network (LAN)/ Wide Area Network (WAN)/ Wireless (e.g., Wi-Fi) adapter, graphics adapter, disk controller, input/output adapter may also be used in addition to or in place of the hardware depicted. The depicted example is provided for the purpose of explanation only and is not meant to imply architectural limitations with respect to present disclosure.
[00084] FIG. 9 is a flow diagram depicting analysis of a chest X-ray image by the AI-powered unified diagnostic image analysis platform 100, according to an exemplary implementation. At block 902, a chest X-ray image associated with a patient is pre-processed using pre-processing technique. At block 904, it is checked whether a chest X-ray image associated with a patient is a standard image. If the chest X-ray image is an inverted image, the check X-ray image is converted to a standard X-ray image. The inversion of the chest X-ray image ensures that the X-ray image conform to a standardized photometric orientation. This prevents diagnostic errors due to polarity mismatch and maintaining reliability across diverse imaging modalities.

[00085] At block 906, presence of lungs in the chest X-ray image is determined using a trained lung detection model. For example, the presence of lungs in the chest X-ray image is determined using a RT-DETR-based object detection model. If the lungs are detected in the chest X-ray image, the chest X-ray image is used to substantially simultaneously perform multiple operations using trained artificial intelligence models as shown in blocks 910. Otherwise, at block 908, the chest X-ray image is rejected.

[00086] At block 910, an abnormality in the lungs is detected by applying the chest X-ray image to a trained abnormality detection model. For example, the abnormality in the lungs is detected using a trained RT-DETR model. The RT-DETR model may be trained to detect four abnormality types: lung nodules, opacity, rib fractures, cardiomegaly.

[00087] At block 912, probability of the individual suffering from Tuberculosis (TB) is determined by applying the chest X-ray image to a trained condition classification model. The trained condition classification model may be a trained multi-layer perceptron model configured to detect tuberculosis in lungs based on the chest X-ray image. For example, the probability of the individual suffering from tuberculosis is indicated as high if consolidation is detected in the lung region.

[00088] At block 914, segmentation of different organs in the chest X-ray is performed using a trained organ segmentation model. A plurality of organ segmentation masks such as lung mask, and heart mask are generated as a result of organ segmentation. At block 916, bone suppressed images are generated by attenuating bony structures (rib and clavicle structures) in the chest X-ray image using a trained bone suppression model. For example, a trained bone suppression model may be an auto-encoder model which generates bone suppressed images by attenuating bony structures in the chest X-ray image while preserving pathological structures including abnormalities which are hidden under the bony structures. Advantageously, bone suppressed views improve visibility of lung parenchyma, nodules, opacities, and subtle consolidations.

[00089] At block 918, abnormality scores including TB score are computed based on the outcome of the blocks 910 and 912. At block 920, a heatmap is generated using feature maps of the detected abnormalities from the block 910. The heatmap overlays provide transparent, evidence-backed reasoning by highlighting spatial regions associated with detected abnormalities. In some embodiments, heatmaps are anatomically constrained using organ segmentation masks from block 914, ensuring clinically meaningful explainability and improved regulatory acceptability.

[00090] At block 922, cardiothoracic ratio (CTR) associated with the patient is computed based on the outcome of the blocks 906 and 916. The CTR is ratio of heart width to chest width. The CTR view highlights cardiac and thoracic boundaries, enabling objective cardiomegaly assessment. CTR serves as a quantitative clinical indicator that provides explainability to the detected abnormality (e.g., cardiomegaly) and supports longitudinal comparison.

[00091] In this manner, the unified diagnostic image analysis platform 100 generates clinically meaningful transformed image views from an original chest X-ray.

[00092] FIG. 10A is a user interface view depicting a medical image 1002 with abnormalities detected by the unified diagnostic image analysis platform 100, according to one embodiment. Specifically, FIG. 10A illustrates a unified diagnostic image analysis interface 1000 displaying a medical image 1002 of a chest of an individual. The medical image 1002 shows an image of a chest of the individual highlighting the detected abnormalities such as cardiomegaly and opacity in the chest via bounding box 1004 and bounding box 1006 respectively. The annotated medical image 1002 provide clinically meaningful explainability and improved regulatory acceptability.

[00093] FIG. 10B is a user interface view depicting a heatmap 1002 generated by the unified diagnostic image analysis platform 100, according to one embodiment. Specifically, FIG. 10B illustrates a unified diagnostic image analysis interface 1050 displaying a heatmap 1008 of a chest of an individual. The heatmap view 1008 shows an image of the chest of the individual highlighting spatial regions associated with detected abnormalities via heatmap activation patterns 1010. The heatmap view 1008 provides clinically meaningful explainability and improved regulatory acceptability.

[00094] The present disclosure provides end-to-end AI-powered medical image analysis platform for assisting radiologists in diagnosing conditions in real-time. The present disclosure provides unified radiological diagnostic ecosystem combining cloud computing, offline edge computing, mobile based access addressing real-world infrastructure constraints across diverse healthcare environments. The platform provides a multi-model deep-learning pipeline for abnormality detection, disease classification, organ segmentation with explainable output, multi-layer workflow system, multi-format reporting, and hybrid deployment model capable of functioning in advanced hospitals as well as resource constrained environment. For example, the unified diagnostic image analysis platform can be deployed as an offline inference box in a resource constrained environment, thereby providing AI-powered diagnostics without internet access.

[00095] The unified diagnostic image analysis platform provides a full radiology workflow automation through DICOM ingestion, AI inference, Interactive annotation tools, disease progression comparison, multi-format reporting, and comprehensive dashboard with real-time operational, diagnostic, and clinical insights. The unified diagnostic image analysis platform provides AI based abnormality and disease detection, organ segmentation, and reporting with a turnaround time of less than 30 seconds per scan. The unified diagnostic image analysis platform provides data-driven, consistent and explainable outputs such as bounding boxes, segmentation masks, heatmaps, bone supressed images, etc. The unified diagnostic image analysis platform ensures low-latency diagnostics, scalable processing, secure storage, and intuitive clinical workflows.

[00096] The above-described examples are for the purpose of illustration. Although the above examples have been described in conjunction with example implementations thereof, numerous modifications may be possible without materially departing from the teachings of the subject matter described herein. Other substitutions, modifications, and changes may be made without departing from the spirit of the subject matter. Also, the features disclosed in this specification (including any accompanying claims, abstract, and drawings), and any method or process so disclosed, may be combined in any combination, except combinations where some of such features are mutually exclusive.

[00097] The terms “include,” “have,” and variations thereof, as used herein, have the same meaning as the term “comprise” or appropriate variation thereof. Furthermore, the term “based on”, as used herein, means “based at least in part on.” Thus, a feature that is described as based on some stimulus can be based on the stimulus or a combination of stimuli including the stimulus. In addition, the terms “first” and “second” are used to identify individual elements and may not meant to designate an order or number of those elements.

[00098] The present description has been shown and described with reference to the foregoing examples. It is understood, however, that other forms, details, and examples can be made without departing from the spirit and scope of the present subject matter that is defined in the following claims. , Claims:WE CLAIM:

1. A method of diagnostic imaging management using an artificial intelligence (AI) powered unified diagnostic image analysis platform, comprising:
obtaining at least one medical image associated with an individual from a medical image database, wherein the at least one medical image corresponds to at least one body part of the individual;
pre-processing, via the unified diagnostic image analysis platform, the medical image associated with the individual; and
performing, via the unified diagnostic image analysis platform, a plurality of operations by applying the pre-processed medical image to a plurality of trained artificial intelligence models, wherein the plurality of operations comprises:
detecting one or more abnormalities associated with the body part by applying the pre-processed medical image to a trained abnormality detection model;
classifying a condition associated with the individual by applying the pre-processed medical image to a trained condition classification model; and
generating at least one explainability output based on at least one of the determined abnormalities and the classified condition associated with the individual.

2. The method as claimed in claim 1, wherein performing, via the unified diagnostic image analysis platform, the plurality of operations by applying the pre-processed medical image to the plurality of trained artificial intelligence models, comprises:
substantially simultaneously performing, via the unified diagnostic image analysis platform, the plurality of operations by applying the pre-processed medical image to the trained artificial intelligence models.

3. The method as claimed in claim 1, wherein performing, via the unified diagnostic image analysis platform, the plurality of operations comprises:
determining whether the pre-processed medical image associated with the individual is an inverted medical image; and
generating a standard medical image using the pre-processed medical image if the medical image is the inverted medical image.

4. The method as claimed in claim 1, wherein performing, via the unified diagnostic image analysis platform, the plurality of operations comprises:
detecting presence of the body part in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model.

5. The method as claimed in claim 1, wherein performing, via the unified diagnostic image analysis platform, the plurality of operations comprises:
performing segmentation of the organs in the pre-processed medical image by applying the pre-processed medical image to a trained organ segmentation model.

6. The method as claimed in claim 1, wherein performing, via the unified diagnostic image analysis platform, the plurality of operations comprises:
generating a bone-suppressed medical image of the individual by applying the pre-processed medical image to a trained bone suppression model.

7. The method as claimed in claim 1, wherein the plurality of operations comprises:
generating a heatmap representing heatmap activation patterns corresponding to the detected abnormalities associated with the body part.

8. An apparatus for diagnostic imaging management, comprising:
a processing unit; and
a memory unit communicatively coupled to the processing unit, wherein the memory unit comprises an artificial intelligence (AI) powered unified diagnostic image analysis platform configured to:
obtain at least one medical image associated with an individual from a medical image database, wherein the at least one medical image corresponds to at least one body part of the individual;
pre-process the medical image associated with the individual; and
perform a plurality of operations by applying the pre-processed medical image to a plurality of trained artificial intelligence models, wherein the one or more operations comprises:
detecting one or more abnormalities associated with the body part by applying the pre-processed medical image to a trained abnormality detection model;
classifying a condition associated with the individual based on the detected abnormalities by applying the pre-processed medical image to a trained condition classification model; and
generating at least one explainability output based on at least one of the detected abnormalities and the classified condition associated with the individual.

9. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
substantially simultaneously perform the plurality of operations by applying the pre-processed medical image to the trained artificial intelligence models.

10. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
determine whether the pre-processed medical image associated with the individual is an inverted medical image; and
generate a standard medical image using the pre-processed medical image if the medical image is the inverted medical image.

11. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
detect presence of the body part in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model.

12. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
perform segmentation of the organs in the pre-processed medical image by applying the pre-processed medical image to a trained organ segmentation model.

13. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
generate a bone-suppressed medical image of the individual by applying the pre-processed medical image to a trained bone suppression model.

14. The apparatus as claimed in claim 8, wherein the unified diagnostic image analysis platform is configured to:
generate a heatmap representing heatmap activation patterns corresponding to the detected abnormalities associated with the body part.

15. A non-transitory computer-readable storage medium, having machine-readable instructions stored therein, that when executed by a processing unit, cause the processing unit to perform method comprising:
obtaining at least one medical image associated with an individual from a medical image database, wherein the at least one medical image corresponds to at least one body part of the individual;
pre-processing the medical image associated with the individual; and
performing a plurality of operations by applying the pre-processed medical image to a plurality of trained artificial intelligence models, wherein the one or more operations comprises:
detecting one or more abnormalities associated with the body part by applying the pre-processed medical image to a trained abnormality detection model;
classifying a condition associated with the individual based on the detected abnormalities by applying the pre-processed medical image to a trained condition classification model; and
generating at least one explainability output based on at least one of the detected abnormalities and the classified condition associated with the individual.

16. The storage medium as claimed in claim 15, wherein the instructions cause the processing unit to perform method comprising:
determining whether the pre-processed medical image associated with the individual is an inverted medical image; and
generating a standard medical image using the pre-processed medical image if the medical image is the inverted medical image.

17. The storage medium as claimed in claim 15, wherein the instructions cause the processing unit to perform method comprising:
detecting presence of the body part in the pre-processed medical image by applying the pre-processed medical image to a trained organ detection model.

18. The storage medium as claimed in claim 15, wherein the instructions cause the processing unit to perform method comprising:
performing segmentation of the organs in the pre-processed medical image by applying the pre-processed medical image to a trained organ segmentation model.

19. The storage medium as claimed in claim 15, wherein the instructions cause the processing unit to perform method comprising:
generate a bone-suppressed medical image of the individual by applying the pre-processed medical image to a trained bone suppression model.

20. The storage medium as claimed in claim 15, wherein the instructions cause the processing unit to perform method comprising:
generate a heatmap representing heatmap activation patterns corresponding to the determined abnormality associated with the body part.