TransFair: Transferring Fairness from Ocular Disease Classification to Progression Prediction

TL;DR

This work tackles the challenge of demographic fairness in ocular disease progression prediction amid limited longitudinal data. It introduces FairEN, a fairness-aware EfficientNet-based classifier, and TransFair, a two-stage knowledge-distillation framework that transfers fairness from classification to progression prediction using a teacher-student setup and $D_{KL}$-based feature alignment. Empirical results on 2D RNFLT maps and 3D OCT data across multiple datasets show that FairEN improves both accuracy and equity in classification, while TransFair yields superior fairness and accuracy in progression prediction compared to strong baselines. The approach demonstrates a practical pathway to achieve fair progression forecasting in ophthalmology and suggesting potential generalization to other diseases with limited longitudinal data.

Abstract

The use of artificial intelligence (AI) in automated disease classification significantly reduces healthcare costs and improves the accessibility of services. However, this transformation has given rise to concerns about the fairness of AI, which disproportionately affects certain groups, particularly patients from underprivileged populations. Recently, a number of methods and large-scale datasets have been proposed to address group performance disparities. Although these methods have shown effectiveness in disease classification tasks, they may fall short in ensuring fair prediction of disease progression, mainly because of limited longitudinal data with diverse demographics available for training a robust and equitable prediction model. In this paper, we introduce TransFair to enhance demographic fairness in progression prediction for ocular diseases. TransFair aims to transfer a fairness-enhanced disease classification model to the task of progression prediction with fairness preserved. Specifically, we train a fair EfficientNet, termed FairEN, equipped with a fairness-aware attention mechanism using extensive data for ocular disease classification. Subsequently, this fair classification model is adapted to a fair progression prediction model through knowledge distillation, which aims to minimize the latent feature distances between the classification and progression prediction models. We evaluate FairEN and TransFair for fairness-enhanced ocular disease classification and progression prediction using both two-dimensional (2D) and 3D retinal images. Extensive experiments and comparisons with models with and without considering fairness learning show that TransFair effectively enhances demographic equity in predicting ocular disease progression.

Figures (7)

• Figure 1: (a) AI-based glaucoma detection is more accessible and affordable compared to traditional clinician diagnoses. Patients can use the outcomes of AI detection as a basis for seeking further diagnosis from clinicians. (b) Longitudinal visual field tests are used to determine the progression of glaucoma. MD: mean deviation - the average of all the values in the visual field map.
• Figure 2: Illustration of the proposed TransFair framework trained in two steps. First, a FairEN model (i.e. teacher) is trained to achieve fairness for classification. Then, the fair classification model guides the training of another FairEN model (i.e. student) to achieve fairness for progression prediction. In TransFair, the teacher and student models take the same inputs but perform classification and progression prediction tasks, respectively. The knowledge distillation aims to minimize the latent feature distances between the teacher and student models.
• Figure 3: The proposed approach for fairness-enhanced progression prediction of ocular disease using retinal images. (a) Train a fair classification model based on the FairEN. (b) Transfer fairness from classification to progression prediction based on the TransFair.
• Figure 4: The glaucoma detection performance using RNFLT maps.
• Figure 5: The glaucoma detection performance using OCT B-scans.