Human-Guided Feature Selection for Accurate Cardiomyocyte Dysfunction Classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Computational Biology

Background:

• Early identification of cardiomyocyte dysfunction is vital for diastolic heart failure (DHF) prognosis.
• Impaired left ventricular relaxation (ILVR) in DHF is linked to inefficient intracellular calcium (Ca2+) handling.
• Analyzing sarcomere length (SL) and calcium kinetics (CK) data is complex for identifying dysfunction.

Purpose of the Study:

• To develop a robust feature selection pipeline for identifying informative features from SL and CK data.
• To create an effective classifier for early detection of cardiomyocyte dysfunction.
• To compare the performance of reduced feature sets against full and PCA-reduced sets.

Main Methods:

• Utilized statistical significance testing, hierarchical clustering, and random forest (RF) classification for feature selection.
• Obtained SL and CK transients from a transgenic mouse model (AAA mice) with ILVR and wild-type controls (NTG).
• Trained and evaluated RF classifiers using full, reduced, and principal component analysis (PCA)-derived feature sets.

Main Results:

• The reduced feature set achieved performance comparable to the full feature set.
• The selected features outperformed the PCA-based approach in classification accuracy.
• The reduced feature set offered improved interpretability by retaining biologically relevant features.

Conclusions:

• A small, curated set of biological features can effectively detect early cardiomyocyte dysfunction.
• The proposed feature selection approach provides precise, interpretable insights for clinical diagnosis.
• This method supports faster diagnosis and intervention decisions for conditions like diastolic dysfunction.