Abstract :

Medical Image Segmentation Is A Crucial Task In Clinical Analysis, But Existing Deep-learning Methods Often Struggle With Ambiguous And Complex Regions. Building On The Gradually Recurrent Network (GREnet).we Propose SiGREnet, A Simplified Gradually Recurrent Network That Integrates Self-supervised Curriculum Learning Directly Into A Single Segmentation Network. By Replacing ConvLSTM With Efficient Temporal Attention (ETA) Layers, SiGREnet Captures Temporal Dependencies Efficiently While Significantly Reducing Computational Complexity And Training Time. A Self-supervised Curriculum Mechanism Dynamically Identifies Hard-to-segment Regions, Allowing The Model To Learn Progressively Without External Supervision. Experiments On Seven Benchmark Datasets-including Dermoscopic, Retinal, Ultrasound, And CT Images - Demonstrate That SiGREnet Achieves Comparable Or Superior Performance To GREnet With Faster Convergence And Lower Computational Overhead, Making It Well-suited For Real-world Clinical Applications. KEYWORDS: SiGREnet, Medical Image Segmentation, 2-D Image Analysis, Simplified Gradually Recurrent Network, Self-Supervised Learning, Curriculum Learning, Deep Learning, Lightweight Architecture, Computational Efficiency, Computer-Aided Diagnosis.

Published:

04-11-2025

Issue:

Vol. 25 No. 11 (2025)

Page Nos:

17-32

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite