Machine learning based classification of intraoperative EMG signals recorded during brain tumor surgeries: a pooled data approach for large-scale analysis and real-time applications.

This study analyzes a publicly available, 7-class iEMG dataset from West China Hospital to prevent nerve damage during brain tumor surgery. Trees, SVM, KNN, Neural Networks, Random Forest, Naive Bayes and 1D-CNN, LSTM, CNN-LSTM models were evaluated. Through data preprocessing, the 80.42% accuracy achieved by Random Forest on original data with a 250 ms window was increased to 97.13% using Bagged Trees on processed data. The study identified 150 ms as the optimal window size for 94.72% accuracy and rapid response. These findings contribute to the literature by establishing the critical balance between speed and accuracy for intraoperative nerve protection.