Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automatic detection and prediction of nAMD activity change in retinal OCT using Siamese networks and Wasserstein Distance for ordinality

Taha Emre, Teresa Araújo, Marzieh Oghbaie, Dmitrii Lachinov, Guilherme Aresta, Hrvoje Bogunović

TL;DR

This work addresses the automated assessment of neovascular AMD activity changes from retinal OCT by proposing two complementary deep learning pipelines evaluated on the MARIO Challenge: SiamRETFound, a Vision Transformer–based Siamese network for longitudinal change detection between visits, and WARIO, an ordinal-aware predictor for 3-month evolution from a single B-scan that leverages MAE pretraining and Wasserstein/EMD losses. SiamRETFound demonstrates strong performance in detecting subtle longitudinal changes, with ensemble predictions surpassing individual models. WARIO tackles severe class imbalance and the ordinal nature of change, using focal loss and EMD loss, and benefits from postprocessing to enforce consistency across OCT volumes, achieving a high leaderboard rank. Together, these methods advance automated OCT-based nAMD management and potentially support timely, personalized anti-VEGF treatment planning.

Abstract

Neovascular age-related macular degeneration (nAMD) is a leading cause of vision loss among older adults, where disease activity detection and progression prediction are critical for nAMD management in terms of timely drug administration and improving patient outcomes. Recent advancements in deep learning offer a promising solution for predicting changes in AMD from optical coherence tomography (OCT) retinal volumes. In this work, we proposed deep learning models for the two tasks of the public MARIO Challenge at MICCAI 2024, designed to detect and forecast changes in nAMD severity with longitudinal retinal OCT. For the first task, we employ a Vision Transformer (ViT) based Siamese Network to detect changes in AMD severity by comparing scan embeddings of a patient from different time points. To train a model to forecast the change after 3 months, we exploit, for the first time, an Earth Mover (Wasserstein) Distance-based loss to harness the ordinal relation within the severity change classes. Both models ranked high on the preliminary leaderboard, demonstrating that their predictive capabilities could facilitate nAMD treatment management.

Automatic detection and prediction of nAMD activity change in retinal OCT using Siamese networks and Wasserstein Distance for ordinality

TL;DR

This work addresses the automated assessment of neovascular AMD activity changes from retinal OCT by proposing two complementary deep learning pipelines evaluated on the MARIO Challenge: SiamRETFound, a Vision Transformer–based Siamese network for longitudinal change detection between visits, and WARIO, an ordinal-aware predictor for 3-month evolution from a single B-scan that leverages MAE pretraining and Wasserstein/EMD losses. SiamRETFound demonstrates strong performance in detecting subtle longitudinal changes, with ensemble predictions surpassing individual models. WARIO tackles severe class imbalance and the ordinal nature of change, using focal loss and EMD loss, and benefits from postprocessing to enforce consistency across OCT volumes, achieving a high leaderboard rank. Together, these methods advance automated OCT-based nAMD management and potentially support timely, personalized anti-VEGF treatment planning.

Abstract

Neovascular age-related macular degeneration (nAMD) is a leading cause of vision loss among older adults, where disease activity detection and progression prediction are critical for nAMD management in terms of timely drug administration and improving patient outcomes. Recent advancements in deep learning offer a promising solution for predicting changes in AMD from optical coherence tomography (OCT) retinal volumes. In this work, we proposed deep learning models for the two tasks of the public MARIO Challenge at MICCAI 2024, designed to detect and forecast changes in nAMD severity with longitudinal retinal OCT. For the first task, we employ a Vision Transformer (ViT) based Siamese Network to detect changes in AMD severity by comparing scan embeddings of a patient from different time points. To train a model to forecast the change after 3 months, we exploit, for the first time, an Earth Mover (Wasserstein) Distance-based loss to harness the ordinal relation within the severity change classes. Both models ranked high on the preliminary leaderboard, demonstrating that their predictive capabilities could facilitate nAMD treatment management.
Paper Structure (21 sections, 2 equations, 6 figures, 1 table)

This paper contains 21 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Longitudinal change detection in retinal OCT
  3. Prediction of nAMD evolution within 3 months on OCT
  4. Contributions
  5. MARIO Challenge
  6. Method
  7. T1: SiamRETFound for longitudinal change detection
  8. Datasets and Evaluation
  9. Training details
  10. Ensembling
  11. T2: WAsserstein distance for RetInal disease Ordinality modeling (WARIO)
  12. Pretraining
  13. Finetuning
  14. Postprocessing (Ensembling)
  15. Datasets and Evaluation
  16. ...and 6 more sections

Figures (6)

  • Figure 1: SiamRETFound approach for longitudinal change detection in retinal OCT.
  • Figure 2: Examples of SiamRETFound predictions (GT: ground truth, P: prediction).
  • Figure 3: Confusion matrices for Longitudinal change detection (MARIO Task 1).
  • Figure 4: Confusion matrices for AMD evolution prediction (MARIO Task 2)
  • Figure 5: Occlusion map sensitivity. Map values are from 0 (blue) to 1 (red). Lower values indicate the occluded region impacted more the final prediction.
  • ...and 1 more figures