Quantification of cardiac capillarization in basement-membrane-immunostained myocardial slices using Segment Anything...

Area of Science:

• Cardiovascular pathology
• Biomedical image analysis
• Computational biology

Background:

• Decreased myocardial capillary density is a key feature in heart disorders.
• Assessing cardiac capillarization typically requires complex double immunostaining and manual image analysis.
• Single basement membrane immunostaining offers a simpler approach but still demands laborious segmentation.

Purpose of the Study:

• To develop an automated image analysis tool, AutoQC, for segmenting cardiomyocytes and capillaries.
• To enable high-throughput and accurate assessment of cardiac capillarization using basement membrane staining.
• To reduce the need for manual labor and expertise in myocardial capillarization analysis.

Main Methods:

• Developed AutoQC using a pre-trained segmentation model (Segment Anything Model, SAM) with prompt engineering.
• Utilized a weakly supervised learning approach requiring only bounding box annotations for training.
• Compared AutoQC's performance against SAM (without prompt engineering) and YOLOv8-Seg for instance segmentation and capillarization assessment.

Main Results:

• AutoQC achieved superior performance in both instance segmentation and capillarization assessment compared to SAM and YOLOv8-Seg.
• The tool enables automatic identification and segmentation of cardiomyocytes and capillaries in basement membrane-stained myocardial slices.
• Weakly supervised training significantly reduced the data annotation workload.

Conclusions:

• AutoQC provides a high-throughput, high-accuracy solution for assessing cardiac capillarization.
• This automated approach simplifies image analysis for basement membrane-stained myocardial tissue.
• AutoQC reduces training requirements and eliminates manual image analysis, accelerating heart disorder research.