A DATA-DRIVEN NORMALIZED ISI ALGORITHM FOR BURST DETECTION LINKS NEURONAL BURST METRICS TO CLINICAL IMPROVEMENT IN PARKINSON'S DISEASE

Pathological burst firing in the subthalamic nucleus (STN) is a hallmark of Parkinson's disease (PD), but robust detection of bursts remains a methodological challenge due to the limitations of classical, parameter-dependent algorithms. We introduce a novel, fully data-driven Normalized ISI Algorithm for burst detection. It employs a probabilistic Gaussian Mixture Model applied to log-transformed and normalized interspike intervals (ISIs), enabling unbiased identification of burst-related activity without prior assumptions or user-defined thresholds. Validated on synthetic data, our method demonstrates superior performance in noisy conditions. Applied to intraoperative STN recordings from PD patients, the algorithm-derived burst metrics showed strong and significant correlations with clinical improvement (UPDRS score change from OFF to ON states). Specifically, the burst spike percentage was inversely correlated, and the mean intra-burst ISI was positively correlated with therapeutic efficacy. These correlations were stronger than those obtained with conventional burst detection methods. Our findings confirm that precise quantification of burst structure, enabled by this objective detection framework, provides a clinically relevant electrophysiological biomarker. The Normalized ISI Algorithm offers a robust tool for linking aberrant neuronal dynamics to motor symptoms, facilitating a deeper understanding of network pathophysiology in movement disorders.