POCA: a CPG signal analysis algorithm using peak-based feature extraction and machine learning.

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Central pattern generator (CPG) research for locomotion relies heavily on neonatal rodent models.
• Biomimetic neural modulation protocols show promise for sustaining fictive locomotor rhythms but require extensive tuning.
• Current CPG signal analysis methods are suboptimal, necessitating automated tools.

Purpose of the Study:

• To introduce the Peak-based Oscillation Classification Algorithm (POCA) for automated CPG signal analysis.
• To develop a novel peak-based feature extraction method for improved classification of locomotor activity.
• To provide a scalable tool for evaluating biomimetic protocols and advancing CPG research.

Main Methods:

• Developed POCA utilizing peak-based feature extraction and machine learning.
• Employed "peak prominence" for thresholding and Support Vector Machine (SVM) with additional peak features.
• Validated POCA on datasets from three independent stimulation protocols.

Main Results:

• The "peak prominence" feature achieved high performance (F1 score: 0.911, accuracy: 0.957).
• SVM incorporating multiple peak features further enhanced classification (F1 score: 0.923, accuracy: 0.966).
• POCA's rhythm characterization results closely matched human expert assessments.

Conclusions:

• POCA offers a robust and scalable solution for CPG signal analysis, facilitating the evaluation of biomimetic protocols.
• The novel peak-based feature extraction framework is adaptable for broader biological oscillation detection.
• This algorithm advances the understanding and analysis of neural control of locomotion.