A mixture-model approach for burst detection in basal ganglia spike trains

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

Abstract Burst activity in basal ganglia spike trains is clinically relevant in Parkinson’s disease and other movement disorders, but burst episodes are heterogeneous: they may differ in duration, intraburst interval structure, relation to pauses or oscillations, and association with motor symptoms. This diversity limits the applicability of universal threshold-based burst detectors to human microelectrode recordings with irregular activity. We introduce LognormISI, a two-stage data-driven method for detecting burst-related regimes by mixture modelling of normalized logarithmic inter-spike interval distributions. The method uses finite log-gamma and Gaussian mixture decomposition to identify statistically separable burst structure without fixed inter-spike interval thresholds. LognormISI was benchmarked against conventional burst detectors using synthetic and clinical data. For physiologically motivated synthetic spike trains (reproducing tonic, pause-burst, tremor-related and beta-related firing patterns of subthalamic nucleus and internal globus pallidus neurons), it consistently favoured high specificity and low false-positive detection over maximal sensitivity, identifying compact, separable burst phenotypes rather than exhaustively labelling all burst-like events. In clinical recordings from Parkinson’s disease patients, several detectors, including LognormISI, have shown the capacity to reveal burst-symptom associations. In summary, LognormISI may provide a reproducible calibration framework for high-confidence burst phenotyping in basal ganglia spike-train analysis along with exploratory analysis of clinical data.

Related topics