Brain Atlas Guided Attention U-Net for White Matter Hyperintensity Segmentation.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• White Matter Hyperintensities (WMH) are common in cerebral small vessel disease (cSVD) and visible on MRI.
• Accurate WMH segmentation is crucial for assessing cSVD burden and clinical outcomes.
• Current segmentation methods often require T1-weighted MRI scans, which are not always available for acute stroke patients.

Purpose of the Study:

• To develop a novel WMH segmentation algorithm that utilizes only FLAIR images and a brain atlas.
• To overcome the limitation of requiring T1-weighted images in existing segmentation approaches.
• To achieve competitive WMH segmentation performance using a more accessible input data combination.

Main Methods:

• Proposed a brain atlas guided attention U-Net (BAGAU-Net) model.
• Designed a dual-path segmentation architecture incorporating a multi-input attention module (MAM) and an attention fusion module (AFM).
• Leveraged spatially-registered white matter (WM) brain atlas with FLAIR images as input.

Main Results:

• The proposed BAGAU-Net achieved competitive WMH segmentation performance.
• Demonstrated effectiveness on two publicly available datasets.
• Outperformed state-of-the-art methods that require T1-weighted images, using only FLAIR and a WM atlas.

Conclusions:

• BAGAU-Net offers an effective solution for WMH segmentation using readily available FLAIR images and a brain atlas.
• This approach simplifies the segmentation process, particularly for acute stroke patients lacking T1-weighted scans.
• The method holds promise for advancing the development and clinical application of WMH segmentation techniques.