Detecting Unforeseen Data Properties with Diffusion Autoencoder Embeddings using Spine MRI data

TL;DR

This paper demonstrates that Diffusion Autoencoder embeddings can effectively uncover biases, data abnormalities, and protocol variations in large-scale spine MRI datasets, improving data quality assessment for medical imaging.

Contribution

It introduces the use of DAE embeddings for bias detection and data quality analysis in medical imaging, outperforming existing generative models in identifying hidden data properties.

Findings

DAE embeddings separate protected variables like sex and age.

t-SNE visualization reveals protocol variations such as head positioning.

Embeddings identify samples challenging for sex prediction models.

Abstract

Deep learning has made significant strides in medical imaging, leveraging the use of large datasets to improve diagnostics and prognostics. However, large datasets often come with inherent errors through subject selection and acquisition. In this paper, we investigate the use of Diffusion Autoencoder (DAE) embeddings for uncovering and understanding data characteristics and biases, including biases for protected variables like sex and data abnormalities indicative of unwanted protocol variations. We use sagittal T2-weighted magnetic resonance (MR) images of the neck, chest, and lumbar region from 11186 German National Cohort (NAKO) participants. We compare DAE embeddings with existing generative models like StyleGAN and Variational Autoencoder. Evaluations on a large-scale dataset consisting of sagittal T2-weighted MR images of three spine regions show that DAE embeddings effectively separate protected variables such as sex and age. Furthermore, we used t-SNE visualization to identify unwanted variations in imaging protocols, revealing differences in head positioning. Our embedding can identify samples where a sex predictor will have issues learning the correct sex. Our findings highlight the potential of using advanced embedding techniques like DAEs to detect data quality issues and biases in medical imaging datasets. Identifying such hidden relations can enhance the reliability and fairness of deep learning models in healthcare applications, ultimately improving patient care and outcomes.