Leveraging 3D Information in Unsupervised Brain MRI Segmentation

TL;DR

This paper introduces a 3D approach to unsupervised brain MRI segmentation using Variational Autoencoders, demonstrating that 3D methods outperform 2D ones in detecting abnormalities like lesions and tumors.

Contribution

The paper proposes a novel 3D UAD framework with a new loss function, improving anomaly detection in brain MRIs over previous 2D approaches.

Findings

3D VAEs outperform 2D VAEs in segmentation accuracy.

A new robust training loss function enhances model performance.

Experiments on multicentric datasets validate the effectiveness of 3D methods.

Abstract

Automatic segmentation of brain abnormalities is challenging, as they vary considerably from one pathology to another. Current methods are supervised and require numerous annotated images for each pathology, a strenuous task. To tackle anatomical variability, Unsupervised Anomaly Detection (UAD) methods are proposed, detecting anomalies as outliers of a healthy model learned using a Variational Autoencoder (VAE). Previous work on UAD adopted a 2D approach, meaning that MRIs are processed as a collection of independent slices. Yet, it does not fully exploit the spatial information contained in MRI. Here, we propose to perform UAD in a 3D fashion and compare 2D and 3D VAEs. As a side contribution, we present a new loss function guarantying a robust training. Learning is performed using a multicentric dataset of healthy brain MRIs, and segmentation performances are estimated on White-Matter Hyperintensities and tumors lesions. Experiments demonstrate the interest of 3D methods which outperform their 2D counterparts.