Multi-Task Deep Learning Framework for Simultaneous Diabetic Retinopathy Grading and Lesion Segmentation Using CNN-Transformer Hybrid Architecture: A Review

Priya Rani Mourya; Adarsh Kumar Soni; Ashish Tamrakar

doi:10.36948/ijfmr.2026.v08i05.87047

Multi-Task Deep Learning Framework for Simultaneous Diabetic Retinopathy Grading and Lesion Segmentation Using CNN-Transformer Hybrid Architecture: A Review

Author(s)	Ms. Priya Rani Mourya, Mr. Adarsh Kumar Soni, Dr. Ashish Tamrakar
Country	India
Abstract	One of the biggest causes of avoidable blindness in working-age populations worldwide is still diabetic retinopathy (DR). Preventing severe visual impairment requires early, accurate diagnosis and exact structural analysis. The substantial clinical link between particular disease symptoms and clinical severity has been ignored by traditional diagnostic paradigms in Deep Learning (DL), which have independently approached either global severity classification (grading) or pixel-level semantic extraction (lesion segmentation). The progressive move toward multi-task learning paradigms that concurrently segment structural lesions and grade DR severity is assessed in this review work. We evaluate how modern frameworks combine Vision Transformers (ViTs), which capture extensive, long-range global contexts, with Convolutional Neural Networks (CNNs), which are excellent at identifying localized structural anomalies. This study describes the state-of-the-art implementations, highlights future development possibilities for clinical deployment, and identifies important gaps in low-quality data processing by analyzing public datasets, architectural hybrids, and optimization techniques. The synergistic potential of multi-task learning is thus investigated in this review paper, which shows how joint optimization of segmentation masks and grading outputs greatly enhances feature representation and clinical explainability. In particular, this study assesses how typical convolutional networks' intrinsic shortcomings in detecting tiny, size-variant micro pathologies might be solved by harmonizing the dualism of global spatial thinking and local lesion sensitivity.
Keywords	Diabetic Retinopathy Grading, Lesion Segmentation, Multi-Task Learning, CNN Transformer Hybrid, Vision Transformer, Deep Learning.
Field	Computer > Data / Information
Published In	Volume 8, Issue 5, September-October 2026
Published On	2026-09-03
DOI	https://doi.org/10.36948/ijfmr.2026.v08i05.87047

View / Download PDF File

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Multi-Task Deep Learning Framework for Simultaneous Diabetic Retinopathy Grading and Lesion Segmentation Using CNN-Transformer Hybrid Architecture: A Review

Share this