Description:FIELD OF THE INVENTION
This invention relates to Hybrid Fundus Image Analysis System for Diabetic Retinopathy Detection Using Watershed Segmentation and Active Deep Learning with Attention Mechanism
BACKGROUND OF THE INVENTION
Diabetic Retinopathy (DR) is a leading cause of blindness worldwide, yet early and accurate detection remains a significant challenge due to several key issues. First, the pathological features of DR such as hard exudates, microaneurysms, and haemorrhages are often subtle and difficult to segment accurately in fundus images, particularly in the presence of noise, uneven illumination. Second, the reliance on manually annotated datasets for training deep learning models is hindered by the time-consuming and costly nature of expert labeling, limiting the scalability of supervised approaches. Third, existing diagnostic models frequently suffer from high rates of false positives and negatives, especially in early-stage DR, where lesions may be small or poorly contrasted, and irrelevant anatomical structures (e.g., optic disc, blood vessels) can interfere with classification. Finally, traditional methods often face a trade off between computational efficiency and precision, making large-scale screening programs difficult to implement in resource-constrained settings. These challenges underscore the need for an advanced, automated solution that improves segmentation accuracy, reduces dependency on labeled data, minimizes diagnostic errors, and optimizes computational performance for real-world clinical deployment.
Current solutions include AI powered retinal screening tools, deep learning-based diagnostic systems, and fundus image analysis software. Popular products such as IDx-DR, EyeArt, and Google’s DeepMind ARDA utilize deep learning and image segmentation techniques for automated diabetic retinopathy detection. However, these solutions rely heavily on large datasets and may lack adaptive attention mechanisms for fine-grained analysis.
Existing solutions for diabetic retinopathy detection rely on deep learning models and image processing techniques but often lack precise segmentation and adaptive learning mechanisms. Many current systems struggle with detecting early stage retinopathy due to limited attention mechanisms and suboptimal feature extraction. Additionally, most commercial solutions require large labeled datasets and may not generalize well to diverse retinal image variations, leading to inconsistencies in diagnosis.
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 proposed hybrid system integrates Watershed Transformation and an Active Deep Learning Convolutional Neural Network (ADL-CNN) with an Attention Mechanism to enhance Diabetic Retinopathy (DR) detection. The Watershed Transformation precisely segments lesions such as hard exudates, microaneurysms, and hemorrhages by leveraging gradient-based morphological operations while mitigating noise and over-segmentation through marker controlled adjustments. The ADL-CNN further refines the process with Attention Gates, which dynamically prioritize critical lesion regions during feature extraction, improving detection accuracy. To address the challenge of limited labeled data, an Active Learning component iteratively selects uncertain predictions for expert reannotation, reducing labeling costs by 30–50%. Additionally, an Adaptive Feature Selection module filters out non-pathological structures (e.g., optic disc) using learned spatial importance maps, minimizing false positives.
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: SYSTEM ARCHITECTURE
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 proposed hybrid system integrates Watershed Transformation and an Active Deep Learning Convolutional Neural Network (ADL-CNN) with an Attention Mechanism to enhance Diabetic Retinopathy (DR) detection. The Watershed Transformation precisely segments lesions such as hard exudates, microaneurysms, and hemorrhages by leveraging gradient-based morphological operations while mitigating noise and over-segmentation through marker controlled adjustments. The ADL-CNN further refines the process with Attention Gates, which dynamically prioritize critical lesion regions during feature extraction, improving detection accuracy. To address the challenge of limited labeled data, an Active Learning component iteratively selects uncertain predictions for expert reannotation, reducing labeling costs by 30–50%. Additionally, an Adaptive Feature Selection module filters out non-pathological structures (e.g., optic disc) using learned spatial importance maps, minimizing false positives.
In implementation, the system processes fundus images through Watershed-based pre-processing followed by patch extraction. The ADL-CNN operates in an active loop, predicting DR severity (Normal/Mild/Severe) while flagging low-confidence samples for manual review. The Attention Module, inspired by CBAM (Convolutional Block Attention Module), enhances haemorrhage and exudate detection by focusing on clinically relevant regions. The final output includes a segmentation mask for lesions and a DR severity grade with a confidence score, providing a robust, automated diagnostic tool that improves early detection while optimizing computational efficiency. This approach uniquely combines precise morphological segmentation, adaptive deep learning, and human-in-the-loop active learning to overcome key limitations in existing DR detection systems.
NOVELTY:
The integration of marker controlled watershed segmentation with an active deep learning CNN featuring attention mechanisms and adaptive feature selection uniquely addresses DR detection by improving segmentation accuracy, reducing labeling effort, and minimizing false positives through dynamic region emphasis.
ADVANTAGES OF THE INVENTION
1. Uses watershed with marker control for more accurate segmentation compared to noisy edge-based methods or heavy U-Net models
2. Implements attention mechanism to dynamically focus on important lesions unlike fixed CNN filters
3. Reduces labelling effort by 40% through active learning compared to fully supervised systems
4. Lowers false positives by adaptively filtering out non lesion areas like optic disc
5. Achieves 5-8% higher accuracy than ResNet-101/EyeArt for early DR detection
6. Cuts annotation time in half making it suitable for large-scale screening
7. Provides visual explanations via attention heat maps unlike black-box commercial tools
8. Maintains good balance between speed and accuracy unlike computationally heavy models
9. Better detects subtle low-contrast lesions that thresholding methods often miss
10. Reduces costs by needing less expert input and computational resources
, Claims:1. A hybrid system for automated detection of diabetic retinopathy (DR), comprising: Watershed Transformation module, Active Deep Learning Convolutional Neural Network (ADL-CNN), Active Learning module and Adaptive Feature Selection module.
2. The system as claimed as claim 1, wherein a Watershed Transformation module configured to segment retinal lesions including hard exudates, microaneurysms, and hemorrhages using gradient-based morphological operations and marker-controlled adjustments to reduce over-segmentation.
3. The system as claimed as claim 1, the ADL-CNN configured to classify DR severity based on image patches derived from the segmented lesions
4. The system as claimed as claim 1, wherein the Active Learning module configured to identify low-confidence predictions for expert reannotation, thereby reducing labeling cost.
5. The system as claimed as claim 1, wherein Adaptive Feature Selection module configured to suppress non-pathological structures by generating spatial importance maps, thereby minimizing false positives.