Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Deep Learning Approaches for Processing Ultra-Widefield Retinal Imaging

Siwon Kim, Wooyung Yun, Jeongbin Oh, Soomok Lee

TL;DR

This paper addresses the need for accurate yet resource-efficient deep learning on ultra-widefield retinal images in low-resource settings. It selects EfficientNet-B0, applies fixed learning rate training, targeted data augmentation, and a multi-model ensemble to achieve strong diagnostic performance for Task 2 (referable DR) and Task 3 (DME) on the UWF4DR dataset. Key contributions include a CPU-friendly training pipeline, an evidence-based backbone selection, and ablation results showing augmentation and ensemble strategies improve AUROC at the cost of inference time, with the model placing 9th in the MICCAI UWF4DR 2024 challenge. The work demonstrates practical, scalable guidance for deploying retinal disease DL systems in clinics with limited hardware, highlighting the balance between accuracy and computational demands.

Abstract

Deep learning has emerged as the predominant solution for classifying medical images. We intend to apply these developments to the ultra-widefield (UWF) retinal imaging dataset. Since UWF images can accurately diagnose various retina diseases, it is very important to clas sify them accurately and prevent them with early treatment. However, processing images manually is time-consuming and labor-intensive, and there are two challenges to automating this process. First, high perfor mance usually requires high computational resources. Artificial intelli gence medical technology is better suited for places with limited medical resources, but using high-performance processing units in such environ ments is challenging. Second, the problem of the accuracy of colour fun dus photography (CFP) methods. In general, the UWF method provides more information for retinal diagnosis than the CFP method, but most of the research has been conducted based on the CFP method. Thus, we demonstrate that these problems can be efficiently addressed in low performance units using methods such as strategic data augmentation and model ensembles, which balance performance and computational re sources while utilizing UWF images.

Efficient Deep Learning Approaches for Processing Ultra-Widefield Retinal Imaging

TL;DR

This paper addresses the need for accurate yet resource-efficient deep learning on ultra-widefield retinal images in low-resource settings. It selects EfficientNet-B0, applies fixed learning rate training, targeted data augmentation, and a multi-model ensemble to achieve strong diagnostic performance for Task 2 (referable DR) and Task 3 (DME) on the UWF4DR dataset. Key contributions include a CPU-friendly training pipeline, an evidence-based backbone selection, and ablation results showing augmentation and ensemble strategies improve AUROC at the cost of inference time, with the model placing 9th in the MICCAI UWF4DR 2024 challenge. The work demonstrates practical, scalable guidance for deploying retinal disease DL systems in clinics with limited hardware, highlighting the balance between accuracy and computational demands.

Abstract

Deep learning has emerged as the predominant solution for classifying medical images. We intend to apply these developments to the ultra-widefield (UWF) retinal imaging dataset. Since UWF images can accurately diagnose various retina diseases, it is very important to clas sify them accurately and prevent them with early treatment. However, processing images manually is time-consuming and labor-intensive, and there are two challenges to automating this process. First, high perfor mance usually requires high computational resources. Artificial intelli gence medical technology is better suited for places with limited medical resources, but using high-performance processing units in such environ ments is challenging. Second, the problem of the accuracy of colour fun dus photography (CFP) methods. In general, the UWF method provides more information for retinal diagnosis than the CFP method, but most of the research has been conducted based on the CFP method. Thus, we demonstrate that these problems can be efficiently addressed in low performance units using methods such as strategic data augmentation and model ensembles, which balance performance and computational re sources while utilizing UWF images.

Paper Structure

This paper contains 17 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Backbone selection
  4. Fine-tuning
  5. Constant learning rate
  6. Data augmentation strategy
  7. Ensemble strategy
  8. Experiments
  9. Datasets and implementation details
  10. UWF Dataset.
  11. Implementation details.
  12. Comparison for backbone selection
  13. Ablation study
  14. Augmentation.
  15. Ensemble.
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Comparison between various retinal imaging datasets. UWF retinal imaging (UWF4DR) and CFP retinal imaging (APTOS 2019, Messidor-2).
  • Figure 2: The overall pipeline of our proposed method. We utilize EfficientNet-B0 and adopt ensemble learning strategy for robustness.