Data-Centric Label Smoothing for Explainable Glaucoma Screening from Eye Fundus Images

TL;DR

The paper addresses glaucoma screening from fundus images under inter-rater variability by introducing Data-Centric Label Smoothing (DC-LS). It encodes annotator disagreement into soft training labels and uses a ResNet50 backbone with 5-fold cross-validation to leverage a large multi-annotator dataset. DC-LS improves sensitivity at 95 percent specificity and reduces Hamming loss on glaucoma feature predictions compared to training on final labels or conventional label smoothing, demonstrating the value of data-centric annotation handling. The work provides a practical, explainable approach for exploiting rich, variably labeled medical imaging data and offers code at github.com/agaldran/justraigs.

Abstract

As current computing capabilities increase, modern machine learning and computer vision system tend to increase in complexity, mostly by means of larger models and advanced optimization strategies. Although often neglected, in many problems there is also much to be gained by considering potential improvements in understanding and better leveraging already-available training data, including annotations. This so-called data-centric approach can lead to substantial performance increases, sometimes beyond what can be achieved by larger models. In this paper we adopt such an approach for the task of justifiable glaucoma screening from retinal images. In particular, we focus on how to combine information from multiple annotators of different skills into a tailored label smoothing scheme that allows us to better employ a large collection of fundus images, instead of discarding samples suffering from inter-rater variability. Internal validation results indicate that our bespoke label smoothing approach surpasses the performance of a standard resnet50 model and also the same model trained with conventional label smoothing techniques, in particular for the multi-label scenario of predicting clinical reasons of glaucoma likelihood in a highly imbalanced screening context. Our code is made available at github.com/agaldran/justraigs .

Figures (1)

• Figure 1: Data-Centric Label Smoothing: depending on the sort of disagreement present on annotations, and the skill of the involved annotators, the degree of smoothing appliced to labels will vary.