White matter hyperintensity and stroke lesion segmentation and differentiation using convolutional neural networks.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• White matter hyperintensities (WMH) are common in aging and small vessel disease, impacting epidemiological studies and clinical trials.
• Manual segmentation of WMH on MRI is time-consuming, costly, and requires expert annotators.
• Distinguishing WMH from other hyperintense lesions like strokes is challenging for automated methods.

Purpose of the Study:

• To develop a fully automated method for segmenting and differentiating WMH from stroke lesions using MRI.
• To introduce a novel convolutional neural network (CNN) architecture, uResNet, for this specific task.
• To evaluate the performance of uResNet against existing algorithms and manual annotations.

Main Methods:

• A fully convolutional CNN architecture, termed uResNet, was designed with analysis and synthesis paths for semantic segmentation.
• The CNN was trained to segment hyperintensities and classify them as either WMH or stroke lesions (cortical, large/small subcortical infarcts).
• Quantitative evaluation involved comparing segmentation overlap with expert annotations and correlation with clinical scores (Fazekas).

Main Results:

• The uResNet CNN achieved superior overlap with manual expert WMH annotations compared to state-of-the-art algorithms.
• WMH volumes segmented by uResNet showed better correlation with the Fazekas visual rating score than other methods.
• Associations between clinical risk factors and WMH volumes from uResNet aligned with those from expert-annotated volumes.

Conclusions:

• The proposed uResNet CNN provides a reliable, automated solution for segmenting and differentiating WMH from stroke lesions on MRI.
• This method offers a more efficient and accurate alternative to manual delineation, with improved clinical relevance.
• The findings support the use of uResNet in epidemiological studies, clinical trials, and understanding small vessel disease.