Description:FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003
Complete Specification
(See section10 and rule13)

1. Title of the Invention: TRANSFORMER-INTEGRATED YOLO WITH LINKED PIXEL EDGE SEGMENTATION FOR LUNG TUMOR DETECTION
2.Applicants: -
SR University India Ananthasagar, Hasanparthy (PO), Warangal-506371, Telangana, India.
3. Inventors:-
Name Nationality Address
Aliya Thaseen Indian School of Computer Science & Artificial Intelligence, SR University, Ananthasagar, Hasanparthy (PO), Warangal-506371, Telangana, India.
Dr. Durgesh Nandan Indian School of Computer Science & Artificial Intelligence, SR University, Ananthasagar, Hasanparthy (PO), Warangal-506371, Telangana, India.
Dr. Sheshikala Martha Indian School of Computer Science & Artificial Intelligence, SR University, Ananthasagar, Hasanparthy (PO), Warangal-506371, Telangana, India.
Preamble to the description:
The following specification particularly describes the invention and the manner in which it is to be performed.

4. DESCRIPTION
FIELD OF THE INVENTION
The present invention means the sphere of medical image processing, i.e., pertain to mechanisms and methodological approaches to earledoning and classifying lung tumors in an early stage. It refers to an integration of both the advanced deep and machine learning algorithms as image segmentation and feature extraction methods. The diagnostics of lung cancer initiated through the invention focuses on accuracy, efficiency, and high-falses. It is particularly applicable in a clinical setting in which a tumor has been detected and defined through the application of the MRI and CT scan images.
BACKGROUND OF THE INVENTION
Lung cancer is a killer cancer that results in cancer death all over the world and with each passing year the statistics of death due to lung cancer are generated under millions. Whereas treatment and imaging modalities, including MRI, CT, PET scan have been built, successful identification of the tumour within the lung at a young age is among the greatest concerns in medical provision. The key phases of the conventional workflows of diagnostic practice most often depend on manual examination of radiology and oncology professionals, and on multi-observer bias, time-consuming and work-consuming, intra-observer bias, and inter-observer bias. Lung tumors are also heterogeneous, without yet having ceased inconvenience of irregularity, of bordering andliness of tissues which make them the more difficult to place into schedules of good typology. Also false positives are prevalent in detecting nodules that result in unneeded tests, biopsies and the development of the psychological discomfort in the patient.
The available computer-aided diagnostic (CAD) systems are also a little bit of a relief because they give the automatic extraction and categorization of features; nevertheless, it has some disadvantages. One drawback of segmentation approaches is that a failure mode occurs with low-contrast images or high-noise images and conventional deep learning layouts tend to over fit because medical data sets having an established ground truth are small in scale. Moreover it is frequently the case that currently available techniques may lack appropriate robustness to address the variability of tumors with regard to size, shape and localization. Another such thorn on the flesh is computational inefficiency: many of the models cannot be used in real time detection and classification because it impacts clinical adoption of this technology.
In order to reduce such issues, the modern research has gone as far as to explore the complexity of hybrid models integrating multiple AI paradigms. Transformer architectures have been compared as statistically better to be able to capture the long-range preferential structures between sequence of images thus very effective in learning features. Simultaneously, object detection networks, including YOLOv5, have already been useful in the localization of tumors in real-time. Alternatively, alternative segmentation methods, including parameter Linked Pixel Edge Segmentation (LPES) give good localization of boundaries, especially with a tumor shaped irregularly. The other step, which is more effective, is the mechanisms of incorporating least correlated weight factors since it mitigates redundant data; more so it is more successful since it has less redundant information.
Together, it will be possible to make the process of diagnosis more accurate, reduce the number of complex operations required to calculate it, and significantly increase the time at its disposal. The invention of this sort can dramatically accelerate the speed/reduce the rates of misdiagnosis/critically enhance the early diagnosis and guide the doctors through the bird-l się to better organize the treatment model. Survival rates the better the better, the better to check towards the truth with which, in the first level, is also fact the inseparable theorist, i.e. to which I vary with the more definitive theorist, of the same research, is itself a constituent that mose the technical knowledge whereof is in themselves (they are) it exists as a decisive influence at the point of view at the general good. Therefore,: we may reach that, there exists extreme allow in new hybridized system that is a precate of another step of transformer, yolo based detection and pixel level classification based optimal feature weight transform the sphere of lung cancer diagnostics.
SUMMARY OF THE INVENTION
The given invention discloses the hybrid lung tumor perceiver structure that uses the transformer facilitated YOLOv5 classification combined with Linked Pixels Edge Segmentation based on the Least Correlated Weight Factor (LPES-LCWF). The system combines the strength of deep learning and machine learning to generate a precision, efficient and real-time detection of lung tumors based on the MRI and CT scan images.
The suggested technique begins by segmenting the lung MRI images using the LPES algorithm that emphasizes more on the pixel-based association and accurate tumor edges even when there are irregular regions. The weight factor mechanism achieves the most optimal on feature selection by reducing redundancies, minimizing false positings and remain with sound feature vectors. After segmentation, features are clustered and reduced to less features based e.g. on ranking. Then the transformer based model including self attention mechanism is trained on the compact features set to learn long range correlations and followed with pairwise connection to yolo v5, further localising and classifying the the Computerised Tomography images and providing benign or malignant tumour. This two-fold ranking and mixed processing allows for the precise identification of tumor and sets constraints on the long processing times.
Experimental testing on benchmark lung cancer data sets demonstrate that the invention is superior to the state-of-the-art for accuracy, precision, recall, and processing efficiency, including Lung-RetinaNet and RAFENet. Besides reducing the processing power, the hybrid approach significantly can provide stable prediction results regardless of the quality of images and demographics of patients. The prototype is being planned to be applied to the actual use, clinical phase application, that would help the radiologists with much faster and reliable diagnostics knowledge. It is scalable across the hospitals and the diagnosis facilities as it is flexible to the various imaging protocols. The invention draws upon the most advanced segmentation ethics, ranking of features and deep learning to design a holistic and novel solution to solicit early onset detection of lung cancer, and hence better patient outcome and reduction of healthcare expenses.
OBJECTIVES OF THE INVENTION
1. The objective is to develop an effective, efficient and real-time mechanism of early detection and clinical classification of lung tumours given the MRI and CT imaging systems.
2. To address state of the art segmentation, feature optimization and hybrid deep learning methods to minimise false positasses and maximise diagnostic accuracy, and
BRIEF DESCRIPTION OF THE DRAWINGS
Fig.1: Depicts Flow Diagram for the proposed invention.
Fig.2: Depicts Lung Cancer diagnostics system.
Fig.3: Depicts YOLO Framework for Lung Tumor Detection.

BRIEF DESCRIPTION OF THE INVENTION
The invention relates to medical imaging, artificial intelligence and computer-aided diagnostics with a particular focus on how the lung tumors are identified and categorized. Lung cancer has been mistaken as one of the most, in the world, fatal cancer and proper early identification is quite inevitable in order to empower patient survival. Manual viewing radiology which is the traditional method of diagnostic works is time consuming, vulnerable to a human factor and most of the time can't identify the tumor that is in its first and vulnerable stage. The proposed invention is a novel and unobvious destructive hybrid construction of superior deep learning and machine learning models for optimizing segmentation, feature identification and classification of lung tumor based on MRIs and CT images.
The presented system is the only one to integrate Linked Pixel Edge Segmentation + Least Correlated Weight Factor (LPES-LCWF) as a finer method of definition of the tumor boundary, and Transformer-integrated YOLOv5 (TYOLOv5) as the means of effective and precise determination of the tumor and its classification. This hybrid framework does not only improve the accuracy of the segmentation, but also provide an efficient features ranking, reducing the computation burden and inclusive of a related real-time practical use in a clinic environment. Combined with segmentation at the pixel level and with deep learning architecture, making use of high level representations, the invention outperforms traditional models in terms of accuracy, reliability and requires lower processing costs.
мәj Huojari (enhancing production) Background and Limitation to the Existing Invented.
Traditionally, diagnosis of lung cancer has been made based on x-ray diagnosis which is defined by chest x-rays and CT scan and MRI scan. Despite the fact that these imaging methodologies give out detailed information in terms of structures, interpretation process heavily relies on proficiency of the radiologist. Separating lung tumors on MRI images manually is a tedious process susceptible to inter-observer errors, thus, providing a non-uniform range of results. Also, the presence of small nodules and tumors in and around complex anatomical structures, e.g., blood vessels and the chest wall are difficult to value with accuracy confidence. The given limitation appears to have been a fundamental factor in generating false reading or false positives and negatives in the existing computer-based systems known as computer-aided diagnostics (CAD).
Complexity Early computer vision and image processing techniques (such as threshold-based segmentation, edge detection and clustering algorithms) offered early tumor detection automation. These approaches were however limited by their inability to generalize to a heterogeneous set of patients and in other imaging conditions. As it progressed machine learning with new deeper learning developed more astute solutions came into place such as convolutional neural networks (CNNs) RetaSetted-based frameworks and U-Distribution frameworks. Although these models were capable of enhancing the performance of detection they were still limited by a number of issues including high computational cost, sensitivity tonoise and inability to operate with abnormal shaped and sized tumors.
Furthermore, deep learning models have many inputs in terms of the training datasets that are usually very large and practically cannot be found in the medical field because of the privacy issues and the cost of gathering the data. Sparsity of this data limits the strengths and the generalisability of traditional AI models. The other essential weakness of the current systems is that they lack efficient mechanisms of choice of features which lead to overlapping, increased rate of computation and low classification.
This invention removes these limitations by a hybrid approach combined with pixel-wise segmentation, the best feature selection based on least-correlated the weight factors and the two feature rankings based on self-attention features of transformers, with the evaluation time required execution of YOLOv5. In such way, it ensures more error-free results, lesser false-positive results, efficiency in processing dimension, as well as better generalisability to variable datasets.
The invention proposes a hybrid multi stage algorithm of lung tumors both in MRI and CT images. It contains the following point of emphasis factors:
A. Series: Linked Pixel Edge-Segmentation (LPES).
The first step involves separating the MRI/CT images of the lung using the Linked Pixel Edge Segmentation technique. This method is interested in pixel based connectivity, so that even minute tumor nodules whose boundaries are irregular or not clearly defined can be correctness captured. LPES provides good continuity of edges, as opposed to the traditional segmentation models, which tends to sample into non tumor areas. The method especially works in identify the presence of the tumor which is close to the complicated parts of the body structure or reaching towards the spiculated tumor.
B. Feature Optimization- Feature scoring by the weighted mean of Least Correlated.
With segmentation the system achieve as Least correlated based Weight Factor, feature extraction and selection. Such a system eliminates the overlapping and unsignificant attributes and keeps the most useful attributed ones to be used for the classification. The LCWF approach is referred to as the most efficient when it comes to estimation of the classification accuracy through striking a balance between more loosely associated features at the expense of other features that offer less important data. This measure further minimise overfitting which occurs using a small sample size for medical imaging datasets.
C. Transformer Feature Processing.
The simplified feature list is later applied to transformer architecture. Contacting features Transformers can learn longer-term dependencies and environmentally-sensitive associations amongst features with their self-access mechanisms. This is very important in medical imaging where slight difference in pixel intensity in one region vs. another can denote important diagnostic data. Transformer writes down the global contextual information, so that it does not ignore minor tumor characteristics.
D. YOLOv5 Real Time Tumor Detection and classification.
High-quality tumor localization and classification after transformer-based processing, YOLOv5 tumor localization is used (You Only Look Once version 5). The fact that YOLOv5 has the highest efficiency in real-time when it comes to this object detecting is quite convenient when it comes to clinical workflows. In terms of bounding boxes, confidence scores, and a probability of the classes (benign or malignant), it allows clinicians to locate the tumor areas easily and with considerable levels of precision. In contrast to classical model detecting regions in the image, YOLOv5 can use the image as one single forwarding request, imanushh ftuthugam transcend much less considerate time on the computation.
SYSTEM WORKFLOW
The complete workflow of the invention involves:
1. Input MRI or CT scan images.
2. Segmentation using LPES for precise boundary identification.
3. Feature extraction and optimization using LCWF.
4. Transformer-based feature processing to capture long-range dependencies.
5. YOLOv5-based real-time tumor localization and classification.
6. Output of tumor classification results, including bounding boxes and malignancy predictions.
In this consolidated workflow, the invention presents an extremely precise, efficient and scalable lung cancer diagnosis method.
BENEFits and appereal uses of the invention.
The offered invention has some important benefit in comparison with the diagnostic systems which are available:
1. High Accuracy and Precision: LPES Segmentation with the integration of transformer-Yolo was found to offer the highest accuracy while detecting and classifying tumors in the lung as compared to currently available Lung-RetinaNet and RAFENet models.
2. Opposed to Reduced False Positives: The LCWF based feature optimization eliminates redundant features which trigger false alarms and the resulting unnecessary costs at the follow-up stage.
3. Live Processing: As the technology has a real time processing feature (YOLOv5) the proposed application can be implemented in hospitals that requires time to be crucial.
4. Skillful Feature Selection Each item in feature set is ranked utilizing dual feature ranking method that classify decision with most informative features boost predictive performance
5. Scalability and Adaptability The popularity of the structure is that it facilitates various imaging modalities such as MRI, CT and PETs scans which leads to the flexibility of the structure across healthcare institutions.
6. Cost-efficiency: The invention cuts down the cost of health care as not much healthcare cost is spent treating the patients and not diagnosing them unnecessary and unnecessary diagnosis, and it is cost-effective for not spending the less health.
7. Clinical Impact: Due to early identification of lung tumors there is direct correlation with the increase in the survival rates. Such a system creates for the clinician a rich decision-support tool that enables a clinician to make interventions and individualized treatment plans earlier.
APPLICATIONS
Specifically: Hospitals and Diagnostic Centers: To screen and diagnose lung cancer (in real-time).
Merit @ the cancer detecting & classifying matter are, the: - Research Institutions: As an inclement to establishment of superior cancer - detecting & classifying schemes.
Telemedicine: this is integration with the system with cloud-based system to improve the remote system healthcare.
Oncology Clinics: As supporting treatment plan by providing a correct capitulate tumor classification result.
The invention is an innovation in the field of computer-aided medical images diagnostics of lung cancer since it integrates segmentation, feature optimization, deep learning and real-time identification into one system. The hybrid solution has promised high degree of accuracy, efficiency as well as the scalability to solve the weak points of the current diagnostic solutions. The invention has high potential of minimizing the burden of lung cancer in the globe due to the potential to dramatically enhance early detection and clinical results.

, Claims:
We Claim:
1. A hybrid lung tumor detection system is composed of Linked Pixel Edge Segmentation with Least Correlated Weight Factor (LPES-LCWF) and transformer built-in YOLOv5 capable of lung tumor detection and classification.
2. The system of claim 1, wherein LPES-LCWF is mandatory that the 2 dimensional boundaries are precisely segmented matching the pixels and removing the redundant features by using correlations based weighting.
3. The system of claim 1, where transformer models use self-attention mechanisms to extract the long-range dependencies to data in medical imaging.
4. The architecture of claim 1, wherein the YOLOv5 is used to localize and classifying lung nodules in real-time on the basis of benign and malignant.
5. The system provided in claim 1, where dual feature ranking is done to enhance accuracy in prediction and reduce the complexity when it comes to calculations.
6. The host of the system of claim 1, which provides a framework flexible to different imaging modalities such as MRI, CT and PET scan.
7. The system of claim 1 wherein integration of LPES-LCWF with Transformer-YOLOv5 has a diagnostics quality greater than that up-to-date lung cancer detection models has.

Dated this 15th September 2025