Wavelet Scattering-Based Classification of Subthalamic Nucleus and Substantia Nigra Microelectrode Recordings During Deep Brain Stimulation Surgery

Deep Brain Stimulation (DBS) surgery offers substantial improvements in quality of life for patients with Parkinson’s disease. It consists of placing a stimulating electrode in the subthalamic nucleus (STN). Accurate and efficient localization of the target is essential because targeting errors may affect patient outcomes and the surgery is performed with the patient awake. Some systems already support neurosurgeons in the STN localization, but reliably separating the STN from the nearby substantia nigra pars reticulata (SNr) remains challenging. This study systematically evaluates the wavelet scattering transform for distinguishing STN from SNr under realistic DBS surgical constraints, where low-frequency information is unavailable and each recording must be classified independently in real time. Varying both the scattering parameters and the model complexity shows how these factors together influence performance. Classifiers ranging from support vector machines to temporal convolutional networks were capable of achieving balanced accuracy of 0.83–0.84 and receiver operating characteristic area under the curve (ROC AUC) values of 0.91–0.93. This suggests that most discriminative information is captured by the wavelet scattering representation itself, resulting in robust performance across models of different complexity. The scattering-based approach achieved consistently higher performance than the evaluated wavelet packet-based baseline and highlights the potential of scattering-based representations for reliable STN localization during DBS surgery. However, the findings were obtained in a single-author, non-preregistered study using data acquired at a single center with a single recording system. Independent multicenter validation will therefore be necessary before broader clinical application can be considered.