Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Surpassing state of the art on AMD area estimation from RGB fundus images through careful selection of U-Net architectures and loss functions for class imbalance

Valentyna Starodub, Mantas Lukoševičius

TL;DR

The study tackles automated AMD area estimation from non-invasive RGB fundus images via semantic segmentation. It builds a U-Net-based framework using pretrained EfficientNet encoders and loss functions designed to mitigate strong class imbalance, evaluating across multiple lesion types against the ADAM challenge benchmark. The authors find that encoder choice (notably EfficientNet-B2) and a weighted binary cross-entropy loss yield the most consistent, state-of-the-art performance in both segmentation and detection, outperforming prior ADAM submissions. The approach delivers a robust, open-source solution with balanced performance across lesion types, supporting non-invasive, scalable AMD assessment in clinical contexts.

Abstract

Age-related macular degeneration (AMD) is one of the leading causes of irreversible vision impairment in people over the age of 60. This research focuses on semantic segmentation for AMD lesion detection in RGB fundus images, a non-invasive and cost-effective imaging technique. The results of the ADAM challenge - the most comprehensive AMD detection from RGB fundus images research competition and open dataset to date - serve as a benchmark for our evaluation. Taking the U-Net connectivity as a base of our framework, we evaluate and compare several approaches to improve the segmentation model's architecture and training pipeline, including pre-processing techniques, encoder (backbone) deep network types of varying complexity, and specialized loss functions to mitigate class imbalances on image and pixel levels. The main outcome of this research is the final configuration of the AMD detection framework, which outperforms all the prior ADAM challenge submissions on the multi-class segmentation of different AMD lesion types in non-invasive RGB fundus images. The source code used to conduct the experiments presented in this paper is made freely available.

Surpassing state of the art on AMD area estimation from RGB fundus images through careful selection of U-Net architectures and loss functions for class imbalance

TL;DR

The study tackles automated AMD area estimation from non-invasive RGB fundus images via semantic segmentation. It builds a U-Net-based framework using pretrained EfficientNet encoders and loss functions designed to mitigate strong class imbalance, evaluating across multiple lesion types against the ADAM challenge benchmark. The authors find that encoder choice (notably EfficientNet-B2) and a weighted binary cross-entropy loss yield the most consistent, state-of-the-art performance in both segmentation and detection, outperforming prior ADAM submissions. The approach delivers a robust, open-source solution with balanced performance across lesion types, supporting non-invasive, scalable AMD assessment in clinical contexts.

Abstract

Age-related macular degeneration (AMD) is one of the leading causes of irreversible vision impairment in people over the age of 60. This research focuses on semantic segmentation for AMD lesion detection in RGB fundus images, a non-invasive and cost-effective imaging technique. The results of the ADAM challenge - the most comprehensive AMD detection from RGB fundus images research competition and open dataset to date - serve as a benchmark for our evaluation. Taking the U-Net connectivity as a base of our framework, we evaluate and compare several approaches to improve the segmentation model's architecture and training pipeline, including pre-processing techniques, encoder (backbone) deep network types of varying complexity, and specialized loss functions to mitigate class imbalances on image and pixel levels. The main outcome of this research is the final configuration of the AMD detection framework, which outperforms all the prior ADAM challenge submissions on the multi-class segmentation of different AMD lesion types in non-invasive RGB fundus images. The source code used to conduct the experiments presented in this paper is made freely available.

Paper Structure

This paper contains 17 sections, 5 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Age-related macular degeneration
  4. Detection of eye diseases using RGB fundus images
  5. Detection of age-related macular degeneration using RGB fundus images
  6. Methods
  7. Semantic segmentation and detection
  8. Data preprocessing and loading pipeline
  9. Model architecture and initialization
  10. Loss functions for semantic segmentation
  11. Training and evaluation pipeline
  12. Results
  13. Evaluation of the encoder choice impact
  14. Evaluation of the loss function impact
  15. Final configuration
  16. ...and 2 more sections

Figures (9)

  • Figure 1: A healthy retina (left) and an AMD-affected retina with yellow drusen (right) bhuiyan2014review
  • Figure 2: Average of the masks over different types of lesions for ADAM dataset (the first five from the left)
  • Figure 3: Input image and multi-channel ground truth mask
  • Figure 4: Original and modified in data loading input image and ground truth mask
  • Figure 5: Input image, ground truth mask, output of segmentation model and its binarization (threshold = 0.5)
  • ...and 4 more figures