Deep Learning-Based Approaches for Ciliary Muscle Segmentation and Biomarker Extraction

EasyChair Preprint 13710

Abstract

This paper highlights our recently published work that involves the application of deep learning techniques to perform the segmentation of the ciliary muscle in Anterior Segment Optical Coherence Tomography (AS-OCT) images. The ciliary muscle is vital for various anterior segment of the eye functions, including intraocular pressure regulation and lens shape maintenance. To advance research, we propose a fully automatic method for segmenting and measuring ciliary muscle biomarkers in 6 mm and 16 mm scan depths, commonly used in clinical analysis. Our approach ensures repeatable and immediate results through thorough exploration of artificial intelligence approaches combining different network architectures, encoders, data augmentation and transfer learning strategies. Additionally, we extract relevant biomarkers, aiding in diagnoses and monitoring of ocular diseases such as glaucoma, myopia, and presbyopia, and facilitating the development of new therapeutic strategies. With high accuracy values (0.9665 ± 0.1280 and 0.9772 ± 0.0873 for the best 6 mm and 16 mm combinations, respectively), our system provides clinicians and researchers with a valuable, automatic tool for ciliary muscle segmentation and analysis in AS-OCT images.

Keyphrases: AS-OCT, CAD system, Ciliary Muscle, Segmentation, biomarkers, deep learning

@booklet{EasyChair:13710,
  author    = {Elena Goyanes and Joaquim de Moura and José Ignacio Fernández-Vigo and José Ángel Fernández-Vigo and Jorge Novo and Marcos Ortega},
  title     = {Deep Learning-Based Approaches for Ciliary Muscle Segmentation and Biomarker Extraction},
  howpublished = {EasyChair Preprint 13710},
  year      = {EasyChair, 2024}}