Cerebral small vessel disease lesion segmentation methods: A systematic review

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Cerebral small vessel disease (CSVD) involves brain lesions detectable via MRI, impacting brain health and dementia risk.
• Manual segmentation of these lesions (WMH, CMB, PVS, lacunes, RSSI) is time-consuming for research.
• Automated machine learning methods are being developed for efficient and accurate CSVD lesion segmentation.

Purpose of the Study:

• To systematically review novel automated methods for segmenting CSVD lesions developed in the last decade.
• To assess the validation of these methods on populations at risk for CSVD.
• To identify publicly available tools and code for CSVD lesion segmentation.

Main Methods:

• Systematic literature search on Web of Science and PubMed.
• Screening and full-text review of 2764 retrieved studies.
• Categorization of automated methods by segmented CSVD lesion type (WMH, CMB, PVS, lacunes, RSSI).

Main Results:

• 89 studies met inclusion criteria, focusing on automated segmentation of CSVD lesions.
• Most methods targeted white matter hyperintensities (59 studies), followed by cerebral microbleeds (23 studies).
• Fewer automated tools were available for perivascular spaces, lacunes, and recent small subcortical infarcts.
• Code or pre-trained models were available for 30 studies, primarily for WMH and CMB segmentation.

Conclusions:

• High-quality automated tools are readily available for white matter hyperintensity segmentation in CSVD.
• There is a need for more validated automated tools for segmenting other CSVD lesion types.
• The development of automated segmentation tools shows promise for advancing CSVD research and dementia prevention.