Video-based machine learning models for predicting deep brain stimulation outcomes in Parkinson's disease patients

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Current Parkinson's disease (PD) deep brain stimulation (DBS) candidate screening relies on subjective levodopa challenge tests (LCT).
• Subjective clinical scales limit the predictive accuracy of LCT for postoperative motor outcomes in PD patients undergoing DBS.
• Objective, quantitative metrics are needed to improve the prediction of DBS efficacy.

Purpose of the Study:

• To develop and validate video-based machine learning models for objective motor assessment during LCT in PD patients.
• To predict binary (DBS+ / DBS-) and ternary (DBS++ / DBS+ / DBS-) postoperative motor outcomes using kinematic data.
• To enhance preoperative screening and patient selection for DBS surgery in Parkinson's disease.

Main Methods:

• Seventy Parkinson's disease patients undergoing DBS surgery were included.
• Video recordings of preoperative levodopa challenge tests (LCT) were analyzed using validated motor assessment software.
• Objective kinematic metrics (velocity, amplitude, stability) were extracted and used to train machine learning models (LDA) for binary and ternary outcome classification.

Main Results:

• Linear Discriminant Analysis (LDA) achieved an F1 score of 0.87 for binary classification (DBS+ / DBS-) and a weighted F1 score of 0.67 for ternary classification (efficacy stratification).
• Models incorporating video-derived kinematic features outperformed baseline models using only conventional clinical predictors.
• Velocity-driven kinematic domains were key predictors, with axial parameters and asymmetric levodopa responses aiding in outcome stratification.

Conclusions:

• Video-based machine learning analysis of LCT provides objective, quantitative motor-responsiveness profiles for Parkinson's disease patients.
• This approach significantly improves the prediction of deep brain stimulation (DBS) efficacy compared to traditional methods.
• Objective LCT analysis can serve as a valuable complementary tool for data-driven patient selection and personalized surgical consultations.