Deep learning-based classification of the capillary ultrastructure in human skeletal muscles

Area of Science:

• Utilizes advanced deep learning techniques, specifically convolutional neural networks (CNNs), for biological image analysis.
• Focuses on the intersection of human skeletal muscle physiology, pathology, and computational methods.

Background:

• Capillary ultrastructure in human skeletal muscle is susceptible to changes caused by systemic conditions like diabetes and hypertension.
• Traditional morphometric analysis of transmission electron microscopy (TEM) images quantifies capillary structure.
• Deep learning offers a novel approach to identify complex patterns in biological data.

Purpose of the Study:

• To train a convolutional neural network (CNN) to detect morphometric differences in muscle capillaries between healthy individuals and patients with systemic pathologies.
• To utilize a hypothesis-generating approach for novel insights into capillary alterations.

Main Methods:

• A retrospective study analyzed 1810 electron micrographs of human skeletal muscle capillaries from 70 participants (healthy controls vs. patients with diabetes mellitus, arterial hypertension, or peripheral arterial disease).
• A pre-trained open-access CNN (ResNet101) was employed to discriminate between capillary images from the two groups.
• The CNN's performance was compared against manual quantitative morphometric analysis, including basement membrane (BM) thickness.

Main Results:

• Manual analysis showed basement membrane (BM) thickness had the highest diagnostic accuracy among morphometric indicators (AUC: 0.657 ± 0.050).
• The best-performing CNN achieved a diagnostic accuracy of 79% (sensitivity 93%, specificity 92%), significantly outperforming BM thickness analysis (p < 0.001).
• The CNN's predictive morphology primarily highlighted pericyte debridement.

Conclusions:

• The hypothesis-generating CNN approach demonstrated superior accuracy in distinguishing between capillaries of healthy individuals and those with systemic pathologies compared to conventional morphometric analysis.
• This deep learning method offers a powerful tool for uncovering subtle ultrastructural differences and generating new research hypotheses in vascular pathology.