Design and implementation of an algorithm for theautomated analysis of Age-Related Macular Degeneration biomarkers on Optical Coherence Tomography

• Álvaro Caballero Sastre ^[1] ; Alberto J. Beltrán Carrero ^[1] ; Javier Torresano Rodríguez ^[2] ; Esther Santos Vicente ^[1] ; María J. Aparicio Hernández-Lastras ^[1] ; María J. Ledesma-Carbayo ^[1] ; Juan-J. Gómez- Valverde ^[1]
  1. [1] Universidad Politécnica de Madrid

     Universidad Politécnica de Madrid

     Madrid, España

     
  2. [2] Hospital Universitario Gregorio Marañón
• Localización: CASEIB 2024. Libro de Actas del XLII Congreso Anual de la Sociedad Española de Ingeniería Biomédica, 2024, ISBN 978-84-09-67332-2, págs. 333-336
• Idioma: inglés
• Enlaces
  • Texto Completo Libro
• Resumen

  • Age-related macular degeneration (AMD) is the leading cause of irreversible vision loss in the elderly in developed countries. This study evaluates advanced deep learning models, including nnUNet, U-Mamba, and MedNeXt, for segmenting key AMD biomarkers such as intraretinal fluid (IRF), subretinal fluid (SRF), and pigment epithelial detachment (PED) in OCT images from a real-world clinical dataset comprising 1,332 OCT B-scans. We compared 2D, 2.5D, and 3D architectures and assessed the impact of incorporating additional biomarkers, such as tubular structures and epiretinal membranes. Our results show that 2D models consistently outperform 3D models, likely due to the nature of the radial scan in our dataset. Specifically, MedNeXt 2D achieved the highest performance, with Dice scores of 0.789 for IRF, 0.691 for SRF, and 0.802 for PED. These findings suggest that leveraging advanced 2D deep learning models can significantly enhance the accuracy of AMD monitoring and treatment, offering substantial improvements in clinical practice.