Intrinsic Bias Identification on Medical Image Datasets

TL;DR

This paper introduces a novel framework for identifying intrinsic biases in medical image datasets, addressing a critical challenge in ensuring the reliability and generalizability of machine learning models in healthcare.

Contribution

It proposes a new bias identification framework with KlotskiNet and Bias Discriminant Direction Analysis, enabling detection of implicit dataset biases that were previously difficult to identify.

Findings

Effective bias attribute discovery on three datasets

Framework outperforms existing bias detection methods

Enhances model generalizability by identifying dataset biases

Abstract

Machine learning based medical image analysis highly depends on datasets. Biases in the dataset can be learned by the model and degrade the generalizability of the applications. There are studies on debiased models. However, scientists and practitioners are difficult to identify implicit biases in the datasets, which causes lack of reliable unbias test datasets to valid models. To tackle this issue, we first define the data intrinsic bias attribute, and then propose a novel bias identification framework for medical image datasets. The framework contains two major components, KlotskiNet and Bias Discriminant Direction Analysis(bdda), where KlostkiNet is to build the mapping which makes backgrounds to distinguish positive and negative samples and bdda provides a theoretical solution on determining bias attributes. Experimental results on three datasets show the effectiveness of the bias attributes discovered by the framework.