NeurostimML: a machine learning model for predicting neurostimulation-induced tissue damage.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Predicting electrical stimulation-induced neural tissue damage is crucial for safe neuromodulation.
• Existing methods rely on limited stimulation parameters, reducing prediction accuracy.
• A more comprehensive approach is needed to enhance the reliability of tissue damage prediction.

Purpose of the Study:

• To develop a machine learning (ML) model for more reliable prediction of electrical stimulation-induced neural tissue damage.
• To incorporate a wider range of stimulation parameters into the predictive model.
• To compare the ML model's performance against the traditional Shannon equation.

Main Methods:

• Compiled a database of 387 stimulation parameter combinations from 58 studies spanning 47 years.
• Utilized ordinal encoding and random forest for feature selection.
• Investigated four ML models (Logistic Regression, K-nearest Neighbor, Random Forest, Multilayer Perceptron) for classification and compared them to the Shannon equation.

Main Results:

• The Random Forest model selected 12 key features, including waveform shape, pulse width, frequency, and various charge and current densities.
• The Shannon equation achieved 63.9% accuracy.
• The Random Forest algorithm demonstrated a significantly higher accuracy of 88.3% in predicting tissue damage.

Conclusions:

• The developed Random Forest model offers a robust and accurate method for predicting neural tissue damage from electrical stimulation.
• This ML-driven approach facilitates informed decision-making for neuromodulation parameter selection in research and clinical settings.
• This study pioneers the use of ML for predicting stimulation-induced neural tissue damage, paving the way for ML-driven neurostimulation.