2D U-NET Deep Learning Architecture for Atherosclerotic Lesion Segmentation and Quantification

A 2D U-NET model quantifies Oil Red O-stained atherosclerotic lesions in mouse aortic root cross-sections with an F1 score of 0.82.

• Why it matters: Manual morphometric measurement on many histology sections is slow and labor-intensive, limiting throughput for experimental atherosclerosis studies in mice.
• What they did: The model was trained on 1,533 Oil Red O-stained aortic root images and tested on 510, with lesions manually segmented in 2,553 images from genetically diverse F2 Apoe knockout mice fed a Western diet for 12 weeks.
• The result: On test images, it matched manual measurements with a Jaccard score of 0.72 and correlation coefficient of 0.98, while reducing per-image measurement time from ~5 minutes to ~3.35 seconds (98.9% faster).

AI summary of the abstract · may contain errors · not medical advice · read the paper.