Basic Science and Pathogenesis.

Area of Science:

• Neuropathology
• Computational Pathology
• Medical Imaging Analysis

Background:

• Current brain arteriolosclerosis assessment relies on subjective, semi-quantitative scales from manual histological examination.
• Manual methods exhibit limited inter-rater reliability and lack precise quantitative metrics.
• Region-specific, high-resolution analysis is often impractical manually.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based algorithm for automated morphometric analysis of arteriolosclerotic blood vessels.
• To introduce Arteriolosclerosis Segmentation (ArtS) and Sclerotic Index and Thickness Extractor (SITE) for quantitative assessment on whole slide images (WSIs).

Main Methods:

• Digitized hematoxylin and eosin-stained brain WSIs from three brain banks.
• Trained three ML models within ArtS for blood vessel detection, arteriolosclerosis classification, and vessel wall/lumen segmentation.
• Utilized the SITE tool to extract sclerotic indices and wall thicknesses from segmented vessels.

Main Results:

• ArtS demonstrated robust performance in blood vessel detection (AUC-ROC up to 0.79) and arteriolosclerosis classification (accuracy up to 0.94).
• Arteriolosclerotic vessel segmentation achieved Dice scores up to 0.73 and AUC-ROC up to 0.92.
• SITE successfully calculated quantitative metrics comparable to expert evaluations.

Conclusions:

• The ArtS and SITE algorithm shows significant potential for quantitative morphometric analysis of arteriolosclerosis.
• These ML tools can enhance current human-based assessment methods for vascular pathology.
• Further optimization could lead to a valuable assistive tool for neuropathologists.