Microelectrode Signals Provide Information Beyond Stereotactic Depth for Subthalamic Nucleus Classification: A Patient-Held-Out Study

During subthalamic nucleus (STN) deep brain stimulation (DBS), microelectrode recording (MER) helps identify the tissue encountered along the planned trajectory. We examined whether MER adds information beyond the electrode's depth relative to the planned target when classifying recordings from patients not used to develop the models.

We analyzed 4650 recordings from 46 patients, assigned to separate development (28 patients), calibration (eight patients), and test (ten patients) groups. Three models used waveform features, target-relative depth, or both. All models and decision thresholds were fixed before testing. Discrimination was measured by the area under the receiver operating characteristic curve (AUROC), averaged equally across patients; 1.0 indicates perfect discrimination and 0.5 indicates chance.

Among 1013 recordings from ten test patients, AUROC was 0.977 (95% CI 0.971-0.983) for signal + depth, 0.957 (95% CI 0.937-0.975) for signal alone, and 0.889 (95% CI 0.848-0.928) for depth alone. Combining signal and depth improved AUROC over depth alone by 0.088 (95% CI 0.051-0.127), with improvement in every patient. At the predefined threshold, sensitivity was 96.8% and specificity was 77.4%. In an exploratory analysis within ±2 mm of the target, AUROC for signal alone remained 0.963, compared with 0.595 for depth alone.

MER provided information beyond target-relative depth for classifying source-labeled STN recordings in previously unseen patients. Combining both inputs gave the highest discrimination. These results support further evaluation of quantitative MER as an adjunct to surgical targeting; they do not establish optimal lead placement or improved clinical outcomes.