Authors:
M. Paul N. Vijaya Kumar, A. Chitra, R. Rajpriya, B. Gayathri, Charlotte Roberts, S. Suman Rajest
Addresses:
Department of Information Science, College of Engineering & Technology, Wollega University, Nekemte, Ethiopia, paulvijji@gmail.com. Department of Computer Science and Applications, St. Peter’s Institute of Higher Education and Research, Chennai, Tamil Nadu, India, prof.chitraa@gmail.com. Department of Computer Science, Chennai National Arts and Science College, Chennai, Tamil Nadu, India, rajpriyatng@gmail.com, bgayathrimca@gmail.com. Australian Graduate School of Engineering (AGSE), UNSW Sydney, Sydney, Australia, charlotteroberts.gs@gmail.com. Department of Research and Development (R&D) & International Student Affairs (ISA), Dhaanish Ahmed College of Engineering, Chennai, Tamil Nadu, India, sumanrajest414@gmail.com.
IRCF is a biomarker for age-related macular degeneration and diabetic macular edoema. IRCF segmentation in OCT images is essential for accurate diagnosis and treatment monitoring. Automated segmentation is needed since manual methods are time-consuming and inconsistent. This study provides an improved U-Net model for OCT IRCF segmentation, which could speed up and improve clinical applications. The model is based on biomedical image segmentation’s U-Net architecture. The skip-connected encoder-decoder structure combines low-level and high-level features while keeping spatial information for accurate IRCF localization. In the training set, several annotated OCT images depict diverse IRCF symptoms. Rotation and flipping improve the dataset’s robustness and generalization. The optimization procedure adjusted hyperparameters like learning rates and reduced overfitting with regularisation. To provide consistent convergence and optimal outcomes, the Adam optimizer and customized learning rate schedule train the model. The U-Net model distinguishes IRCF regions in OCT images, suggesting superior segmentation. Its proven clinical efficacy simplifies eye disease diagnosis and treatment. To establish deep learning’s impact on improving medical imaging for improved ophthalmic treatment, future studies will evaluate its applicability to various retinal biomarkers and across patient cohorts.
Keywords: Intraretinal Cystoid Fluid; Optical Coherence Tomography; U-Net Architecture; Diabetic Macular Edema (DME); Retinal Vein Occlusion (RVO); Applying Regularization; Segmentation Techniques.
Received on: 15/09/2023, Revised on: 22/11/2023, Accepted on: 02/12/2023, Published on: 07/03/2024
AVE Trends in Intelligent Health Letters, 2024 Vol. 1 No. 1, Pages: 38-50