Description:FIELD OF THE INVENTION
The present invention relates to the field of medical image analysis and artificial intelligence. Specifically, it concerns the detection and classification of diabetic retinopathy using deep learning techniques to assist in early diagnosis and improved clinical decision-making.
BACKGROUND OF THE INVENTION
Diabetic retinopathy (DR) is a serious complication of diabetes that affects the retina and can lead to visual loss if not treated in a timely manner. One traditional method for diagnosing DR is the manual examination of the discovered images of the retina by an ophthalmologist. Ophthalmologists consume time, are susceptible to human errors, and remove the effects.
To address these challenges, deep learning techniques have been increasingly used to automate the detection and classification of DR. Folding networks (CNNSs), such as VGG16, Reset, and Ethernet, show promising results in the analysis of retinal images to identify signs of DR, including microbial urine, bleeding, exudates, and neovascularization. However, existing deep learning models face the following challenges: imbalanced data records, inadequate generalization of various population groups, and high demand for arithmetic resources. This study aimed to develop a robust, deep learning-based DR recognition and classification system that improves accuracy, reduces false diagnoses, and improves early detection. The proposed model uses methods such as data expansion, transmission learning, and explainable AI algorithms to ensure reliability and transparency in decision-making. By automating DR diagnosis, this approach supports ophthalmologists, reduces stress in the health system, and promotes early treatment to prevent visual impairment.
Several deep learning solutions have been developed for the detection and classification of diabetic retinopathy (DR). Some notable solutions are as follows:
Google's advanced AI employs different learning models to detect DR and macular edema.
Retinalization: A cloud-based AI tool for detecting retinal diseases using fundus images and ophthalmologists. This was integrated into the ophthalmic workflow.
2. Current Limitations on Solutions
Although existing solutions have made considerable advances, there are still some mistakes. False alarms and false negatives: Current AI models may not classify normal cases as DR in the early stages or recognize DR.
Results and Patents:
The patent, AI-based diabetic retinopathy detection (US10935678B2), focuses on the use of deep learning models for DR classification. The patent Machine Learning for Automated DR Screening (WO2019182149A1) describes an ML-based approach for automated DR diagnosis. 2023) described a CNN and hybrid AAI approach for DR recognition. PhD Classification Patents "
"Automized fundus image analysis AI patent"
"AI screen patent for diabetic eye diseases"
"Explainable AI of ophthalmic patents"
We will continue to impose accuracy, affordability, and interpretability of clinical applications worldwide.
Feature Existing Solutions Proposed Solution
Model Type Uses standard CNNs (VGG16, ResNet, Inception) for classification. Hybrid YOLO-based lesion detection + CNN-based classification for higher accuracy.
Image Processing Requires high-quality images for accurate predictions. Integrates CLAHE & MCW techniques to enhance low-quality images, improving lesion visibility.
Generalization Models struggle with dataset biases and fail to generalize across different populations. Uses domain adaptation & transfer learning to ensure better generalization across diverse datasets.
Explainability Most deep learning models are black-box systems, offering no insight into predictions. Implements Explainable AI (XAI) techniques like Grad-CAM, SHAP, LIME for transparent decision-making.
DR Lesion Detection Limited bounding-box detection; mostly focuses on classification. Uses YOLO for precise lesion localization & severity grading.
Real-Time Screening Mostly hospital-based AI systems requires high computing resources. Deployable via cloud API & mobile-based Edge AI, making DR detection accessible anywhere.
Computational Efficiency Some solutions require high-end GPUs & cloud computing for analysis. Optimized for mobile devices & edge computing for low-cost, real-time analysis.
Scalability Available in specialized hospitals, but not easily accessible to remote areas. Mobile-friendly and cloud-integrated, enabling mass screening in low-resource settings.
Performance on Noisy Data Performance drops with low-resolution or noisy images. Noise reduction & artifact removal techniques ensure robust performance on real-world images.

OBJECTIVES OF THE INVENTION
Main objective of the present invention is to develop an automated system for accurate detection and classification of diabetic retinopathy (DR) using deep learning models trained on retinal fundus images and OCT scans.
Another objective of the present invention is to improve image analysis performance through advanced pre-processing techniques including resizing, normalization, data augmentation, and enhancement using CLAHE and Multichannel Wavelet methods.
Another objective of the present invention is to implement optimized convolutional neural networks (CNNs) such as ResNet-50, VGG16, DenseNet, and YOLOv8 for effective lesion detection and multiclass classification of DR severity.
Another objective of the present invention is to provide explainable AI-based interpretability through techniques like Grad-CAM, SHAP, and LIME, enabling clinicians to visualize the decision-making process of the AI system.
Another objective of the present invention is to ensure accessibility and scalability of the DR screening system through deployment on cloud-based platforms and mobile applications, supporting both clinical and remote healthcare environments.
SUMMARY OF THE INVENTION
This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the invention.
This summary is neither intended to identify key or essential inventive concepts of the invention and nor is it intended for determining the scope of the invention.
To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.
The objective of the proposed solution is to improve the accuracy, accessibility, and interpretability of the detection and classification of diabetic retinopathy (DR) using an advanced deep learning model. Using a folding network (CNNS), transfer learning, and explainable AI (XAI) technology, the system automatically analyzes retinal fund images, recognizes early stages, and categorizes DR-LEVEL (normal, mild, moderate, severe, proliferation, DR).
Herein enclosed a computer-implemented method for automated detection and classification of diabetic retinopathy (DR), comprising the steps of:
collecting retinal fundus images or optical coherence tomography (OCT) scans from public datasets such as Kaggle Aptos, Eyepacs, Messidor-2, and clinical datasets;
pre-processing said images by resizing to a uniform resolution, normalizing pixel values, augmenting data using rotation, flipping, and brightness adjustments, and enhancing image contrast using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Multichannel Wavelet (MCW) techniques;
training a deep learning model selected from a group comprising ResNet-50, VGG16, DenseNet, and YOLOv8, wherein the model is fine-tuned to extract features from said images and classify DR into severity categories using SoftMax activation;
performing model inference and prediction by identifying lesions such as microaneurysms, hemorrhages, and exudates in real-time;
providing model interpretability through Explainable AI techniques including Grad-CAM, SHAP, and LIME to visualize and justify the decision-making process of the deep learning model;
displaying results to the user via a cloud-based API or mobile application and generating a DR diagnosis report indicating the DR severity level and highlighted lesion regions.
The deep learning model comprises a backbone network for residual feature extraction, a YOLO-based viewing head for bounding box localization of DR lesions, and a fully connected layer (FCL) classifier for final categorization.
The DR is classified into five severity categories: normal, mild, moderate, severe, and proliferative diabetic retinopathy.
The system is operable via a cloud-based AI platform for use in hospitals and clinics and via a mobile or edge AI application for remote and affordable DR screening.
The diagnostic accuracy is improved by image quality enhancement through CLAHE and MCW, resulting in robustness against low-quality inputs and reduced false positives.
The output DR diagnosis report comprises bounding box visualizations of detected lesions and a confidence score reflecting the model’s diagnostic certainty.
BRIEF DESCRIPTION OF THE DRAWINGS
The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example, and simply illustrates certain selected embodiments of devices, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
FIGURE 1: flow chart for dr detection and classification
The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION OF THE INVENTION
The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present disclosure as defined by the appended claims.
It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a",” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
In addition, the descriptions of "first", "second", “third”, and the like in the present invention are used for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Thus, features defining "first" and "second" may include at least one of the features, either explicitly or implicitly.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The objective of the proposed solution is to improve the accuracy, accessibility, and interpretability of the detection and classification of diabetic retinopathy (DR) using an advanced deep learning model. Using a folding network (CNNS), transfer learning, and explainable AI (XAI) technology, the system automatically analyzes retinal fund images, recognizes early stages, and categorizes DR-LEVEL (normal, mild, moderate, severe, proliferation, DR).
The System addresses the most important limitations of existing solutions.
Improved accuracy: An optimized CNN architecture (e.g., Yolo, Reset, or Efficient).
Improved generalization: Uses a variety of data records and data enlargement techniques. Ensure interpretability by including Grad-CAM and SHAP or LIME for model declarations.
Reduced dependence on high-quality images Contrast-limited adaptive histogram compensation (CLAHE) and Modified Cross-Wave (MCW) technology.
Providing affordable, scalable solutions via cloud-based or edge AI provisioning on mobile devices.
Herein enclosed a computer-implemented method for automated detection and classification of diabetic retinopathy (DR), comprising the steps of:
collecting retinal fundus images or optical coherence tomography (OCT) scans from public datasets such as Kaggle Aptos, Eyepacs, Messidor-2, and clinical datasets;
pre-processing said images by resizing to a uniform resolution, normalizing pixel values, augmenting data using rotation, flipping, and brightness adjustments, and enhancing image contrast using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Multichannel Wavelet (MCW) techniques;
training a deep learning model selected from a group comprising ResNet-50, VGG16, DenseNet, and YOLOv8, wherein the model is fine-tuned to extract features from said images and classify DR into severity categories using SoftMax activation;
performing model inference and prediction by identifying lesions such as microaneurysms, hemorrhages, and exudates in real-time;
providing model interpretability through Explainable AI techniques including Grad-CAM, SHAP, and LIME to visualize and justify the decision-making process of the deep learning model;
displaying results to the user via a cloud-based API or mobile application and generating a DR diagnosis report indicating the DR severity level and highlighted lesion regions.
The deep learning model comprises a backbone network for residual feature extraction, a YOLO-based viewing head for bounding box localization of DR lesions, and a fully connected layer (FCL) classifier for final categorization.
The DR is classified into five severity categories: normal, mild, moderate, severe, and proliferative diabetic retinopathy.
The system is operable via a cloud-based AI platform for use in hospitals and clinics and via a mobile or edge AI application for remote and affordable DR screening.
The diagnostic accuracy is improved by image quality enhancement through CLAHE and MCW, resulting in robustness against low-quality inputs and reduced false positives.
The output DR diagnosis report comprises bounding box visualizations of detected lesions and a confidence score reflecting the model’s diagnostic certainty.
EXAMPLE 1
BEST METHOD
Implementation Details: How the System Works
Step 1: Data Collection and Pre-Processing
Dataset: Photos of retinal fundus from Kaggle (Aptos, Eyepacs), Messidor-2, and actual clinical datasets.
Preparation Method:
Resize: Stand-up images with uniform size (for example, 640 is 640 pixels).
Normalization: Scaling pixel values for a better contrast.
Grow data: Rotation, flipping, and brightness adjustment.
Imaging Improvements: Application of CLAHE and MCW, highlighting lesions around the DR.
Step 2: Deep Learning-Based DR Recognition and Classification
CNN Model Selection:
Feature Extraction: Use prepared models (ResNet-50, vgg16, Dens Net, yolov8) to extract DR features.
Fine-tuning and sending learning: The model was adapted to the DR dataset by retraining the final level. Multiclass Classification: Category DR Severity Uses SoftMax Activation.
Model Architecture:
Backbone: Efficient network for residual 50/function extraction.
Viewing Head: YOLO-based boundary box recognition of the lesion location.
Classifier: Classification from fully connected layers (FCL) of the DR.
Step 3: Model Inference and Explanation
Real-time detection: Users load retinal images using a cloud-based API or mobile application.
Lesion detection and classification: Models identified, Micro biotic disease, bleeding, exudation
Explainable AI (XAI) for interpretability
Grad-CAM: Emphasizes the area used in the CNN model for decision-making.
SHAP/LIME: supplied feature values for model prediction.
Status Level: Model displays diagnostic security.
Step 4: Ophthalmologist Edition and Decision Support
Final Question:
Several class formations (normal, mild, moderate, severe, and proliferative DR).
Restriction box highlighting the perceived lesions.
Mission:
Cloud-based AI System for hospitals and clinics.
Mobile App/Edge AI For remote screening.
Major Benefits Compared to Existing Solutions
High accuracy: Optimized CNN models were improved. The recognition accuracy was improved, and false alarms were reduced.
Better Interpretability: XAI Techniques. Helps doctors trust AI decisions.
Robust against image quality issues: CLAHE & MCW Technology improve contrast and lesion visibility.
Accessible and affordable: A Mobile-based AI Solution ensuring widespread DR screening.
Scalable for large-scale deployment: Cloud-based and mobile-friendly AI systems enable mass screenings.
NOVELTY:
The newly proposed system for detecting and classifying Diabetic Retinopathy (DR) offers several innovative features that set it apart from current solutions.
1. Hybrid Deep Learning Method for Enhanced Precision: This system integrates YOLO-based lesion detection with CNN-based classification models, such as ResNet-50 and Efficient Net, to achieve accurate grading of DR severity. It employs multi-feature extraction methods to identify subtle DR indicators, such as microaneurysms, hemorrhages, and exudates, more precisely.
2. Advanced Image Processing for Improved Performance Techniques, such as contrast-limited adaptive histogram equalization (CLAHE) and modified cross-wavelet (MCW), enhance the contrast and sharpness of fundus images, making DR features easier to detect. The system also reduces noise and removes artifacts to ensure better performance with low-quality or real-world clinical-image data.
3. Explainable AI (XAI) for Clarity and Reliability: The system uses Grad-CAM, SHAP, and LIME to highlight the regions involved in decision-making, addressing the "black-box" problem of deep learning. It provides confidence scores and visual explanations to aid ophthalmologists in their decision-making.
4. Real-time and Scalable Implementation:
Implementing a cloud-based API and mobile application facilitates the real-time detection of DR in remote locations, enhancing the accessibility and cost-effectiveness of screening. Optimizing Edge AI allows deployment on low-power devices, such as smartphones and edge GPUs, without the need for extensive computational resources. Adaptive Learning for Enhanced Generalization employs domain adaptation methods to boost model performance across various populations and imaging devices. A continuous learning framework progressively refines the model to minimize biases in the datasets. Key Innovative Contributions include the pioneering YOLO + CNN hybrid for DR detection and classification, the integration of CLAHE and MCW for superior image preprocessing, and the use of explainable AI techniques (Grad-CAM, SHAP, LIME) to ensure transparency and build medical trust. The scalable real-time solution through cloud and mobile deployment supports mass screening, whereas the adaptive learning framework enhances performance across diverse datasets. This innovative approach guarantees greater accuracy, improved interpretability, and enhanced accessibility, making DR detection more efficient and reliable for practical clinical applications.
ADVANTAGES OF THE INVENTION
Enhanced Precision: The integration of YOLO-based lesion detection with CNN classification boosts accuracy.
Improved Generalization: Adaptive learning strategies enhance the robustness of the model across various datasets.
Increased Explainability: XAI tools, such as Grad-CAM and SHAP, provide transparency in AI decisions for medical professionals.
Greater Accessibility and Scalability: A cloud-based and mobile-compatible solution facilitates mass screening and serves remote locations.
Effective on Low-Quality Images: Utilizes CLAHE and MCW to improve fundus images, minimizing reliance on high-end cameras.
Conclusion: This approach surpasses current methods in terms of accuracy, interpretability, scalability, and real-time accessibility, making DR detection quicker, more reliable, and broadly applicable.
, Claims:1. A computer-implemented method for automated detection and classification of diabetic retinopathy (DR), comprising the steps of:
a) collecting retinal fundus images or optical coherence tomography (OCT) scans from public datasets such as Kaggle Aptos, Eyepacs, Messidor-2, and clinical datasets;
b) pre-processing said images by resizing to a uniform resolution, normalizing pixel values, augmenting data using rotation, flipping, and brightness adjustments, and enhancing image contrast using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Multichannel Wavelet (MCW) techniques;
c) training a deep learning model selected from a group comprising ResNet-50, VGG16, DenseNet, and YOLOv8, wherein the model is fine-tuned to extract features from said images and classify DR into severity categories using SoftMax activation;
d) performing model inference and prediction by identifying lesions such as microaneurysms, hemorrhages, and exudates in real-time;
e) providing model interpretability through Explainable AI techniques including Grad-CAM, SHAP, and LIME to visualize and justify the decision-making process of the deep learning model;
f) displaying results to the user via a cloud-based API or mobile application and generating a DR diagnosis report indicating the DR severity level and highlighted lesion regions.
2. The method as claimed in claim 1, wherein the deep learning model comprises a backbone network for residual feature extraction, a YOLO-based viewing head for bounding box localization of DR lesions, and a fully connected layer (FCL) classifier for final categorization.
3. The method as claimed in claim 1, wherein the DR is classified into five severity categories: normal, mild, moderate, severe, and proliferative diabetic retinopathy.
4. The method as claimed in claim 1, wherein the system is operable via a cloud-based AI platform for use in hospitals and clinics and via a mobile or edge AI application for remote and affordable DR screening.
5. The method as claimed in claim 1, wherein diagnostic accuracy is improved by image quality enhancement through CLAHE and MCW, resulting in robustness against low-quality inputs and reduced false positives.
6. The method as claimed in claim 1, wherein the output DR diagnosis report comprises bounding box visualizations of detected lesions and a confidence score reflecting the model’s diagnostic certainty.