A Retinal Vessel Segmentation Method Based on the Sharpness-Aware Minimization Model

Sensors, Journal Year: 2024, Volume and Issue: 24(13), P. 4267 - 4267

Published: June 30, 2024

Retinal vessel segmentation is crucial for diagnosing and monitoring various eye diseases such as diabetic retinopathy, glaucoma, hypertension. In this study, we examine how sharpness-aware minimization (SAM) can improve RF-UNet’s generalization performance. RF-UNet a novel model retinal segmentation. We focused our experiments on the digital images extraction (DRIVE) dataset, which benchmark segmentation, test results show that adding SAM to training procedure leads notable improvements. Compared non-SAM (training loss of 0.45709 validation 0.40266), SAM-trained achieved significant reduction in both (0.094225) (0.08053). Furthermore, compared accuracy 0.90169 0.93999), demonstrated higher (0.96225) (0.96821). Additionally, performed better terms sensitivity, specificity, AUC, F1 score, indicating improved unseen data. Our corroborate notion facilitates learning flatter minima, thereby improving generalization, are consistent with other research highlighting advantages advanced optimization methods. With wider implications medical imaging tasks, these imply successfully reduce overfitting enhance robustness models. Prospective avenues encompass verifying vaster more diverse datasets investigating its practical implementation real-world clinical situations.

Language: Английский

Medical image segmentation method based on full perceived dynamic network

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 142, P. 109867 - 109867

Published: Dec. 22, 2024

Language: Английский

Convolutional state space model with multi-window cross-scan to advance the automated diagnosis of skeletal fluorosis

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 103, P. 107439 - 107439

Published: Dec. 28, 2024

Language: Английский