Automated Classification of Store-Operated Calcium Entry Activity and Disease Conditions in Murine Skeletal Muscle...

Area of Science:

• Biomedical imaging
• Machine learning in pathology
• Skeletal muscle physiology

Background:

• Accurate diagnosis of muscular dystrophies relies on identifying tissue pathophysiology.
• Machine learning (ML) offers advanced image assessment capabilities.
• Store-operated calcium entry (SOCE) activity serves as a biomarker for muscle activity and disease.

Purpose of the Study:

• To compare the efficacy of three ML models in classifying mouse skeletal muscle images.
• To assess ML model performance in identifying SOCE activity as an indicator of muscle disease.

Main Methods:

• Immunofluorescent images of muscle fibers from wildtype and calpain-3 null mice were analyzed.
• Images were categorized by SOCE activity and disease status, then split into training, validation, and testing sets.
• Three deep learning models were employed: Convolutional Neural Networks (CNN), EfficientNet, and Support Vector Machines (SVM).

Main Results:

• CNN achieved the highest accuracy (0.91) and F1 score (0.88).
• SVM demonstrated the highest precision (0.92).
• Both CNN and SVM showed comparable performance with an area under the receiver operating characteristic curve of 0.91, with no significant difference (p=0.19).

Conclusions:

• CNN and SVM models are effective for classifying SOCE activity in muscle images.
• These ML models provide scalable, automated solutions for tissue classification in muscle pathologies.
• Future work should involve larger datasets and advanced models like transformers for complex muscle conditions.