Class Unbiasing for Generalization in Medical Diagnosis

TL;DR

This paper addresses class-feature bias in medical diagnosis models, proposing a novel training method that mitigates bias and class imbalance, leading to improved generalization on diverse classes.

Contribution

It introduces a class-wise inequality loss and a class-weighted optimization approach to reduce bias and imbalance simultaneously.

Findings

Effectively mitigates class-feature bias and class imbalance.

Improves model generalization on real-world datasets.

Demonstrates robustness through synthetic experiments.

Abstract

Medical diagnosis might fail due to bias. In this work, we identified class-feature bias, which refers to models' potential reliance on features that are strongly correlated with only a subset of classes, leading to biased performance and poor generalization on other classes. We aim to train a class-unbiased model (Cls-unbias) that mitigates both class imbalance and class-feature bias simultaneously. Specifically, we propose a class-wise inequality loss which promotes equal contributions of classification loss from positive-class and negative-class samples. We propose to optimize a class-wise group distributionally robust optimization objective-a class-weighted training objective that upweights underperforming classes-to enhance the effectiveness of the inequality loss under class imbalance. Through synthetic and real-world datasets, we empirically demonstrate that class-feature bias can negatively impact model performance. Our proposed method effectively mitigates both class-feature bias and class imbalance, thereby improving the model's generalization ability.