View-Disentangled Transformer for Brain Lesion Detection

TL;DR

This paper introduces a view-disentangled transformer that improves brain lesion detection in MRI scans by efficiently capturing 3D correlations through multiple 2D views, leading to more accurate tumor localization.

Contribution

The paper proposes a novel transformer architecture that models 3D brain MRI features as multiple 2D views, enhancing lesion detection accuracy while maintaining computational efficiency.

Findings

Effective detection of brain lesions on MRI datasets.

Improved accuracy over existing methods.

Efficient modeling of 3D correlations in 2D views.

Abstract

Deep neural networks (DNNs) have been widely adopted in brain lesion detection and segmentation. However, locating small lesions in 2D MRI slices is challenging, and requires to balance between the granularity of 3D context aggregation and the computational complexity. In this paper, we propose a novel view-disentangled transformer to enhance the extraction of MRI features for more accurate tumour detection. First, the proposed transformer harvests long-range correlation among different positions in a 3D brain scan. Second, the transformer models a stack of slice features as multiple 2D views and enhance these features view-by-view, which approximately achieves the 3D correlation computing in an efficient way. Third, we deploy the proposed transformer module in a transformer backbone, which can effectively detect the 2D regions surrounding brain lesions. The experimental results show that our proposed view-disentangled transformer performs well for brain lesion detection on a challenging brain MRI dataset.