MCWs (MiCroWire sorter): A new framework for automated and reliable spike sorting in human intracerebral recordings

Efficient and accurate spike sorting is critical for isolating single neurons from extracellular recordings to distinguish neural activity of interest. However, while the electrodes and acquisition systems for non-human electrophysiology have been enhanced over the past decades to enable higher-yield single-neuron detections, those advances have not been translated into human electrophysiology. Single-wire electrodes are still ubiquitously used, and although acquisition systems have augmented their signal-to-noise ratio over the last 15 years, we are still limited by their low electrode count. Moreover, unlike non-human recordings, human recordings often take place in hospitals where different noise sources and subject breaks can compromise the recording quality during experimental sessions. To bridge this gap, this work presents an automatic, open-source spike sorting pipeline that leverages contemporary computational capabilities and is tailored to single-neuron recordings from humans acquired via microwires. The pipeline is implemented in both MATLAB and Python, ensuring accessibility and compatibility across computational environments. Its modular and comprehensive structure supports customization and even opportunities for new developments as per the requirements of the user and the application. One feature is a data-driven automatic module to remove narrow-band interference, besides electrical line noise, which can be an essential tool while recording in clinical settings, particularly for online processing implementations. Following spike detection, the pipeline implements an artifact rejection module that separates waveforms that are unlikely to be associated with actual spikes. Additionally, we introduce a configurable feature-extraction, clustering, and benchmarking framework that not only allows flexibility in employing user-defined or conventional algorithms, such as wavelet transform with superparamagnetic clustering, but can also evaluate multi-method agreement among the different sorters. The pipeline also utilizes established and novel quality metrics to support semiautomatic curation of isolated clusters. Furthermore, we can integrate the customized pipeline with experimental tasks by removing task-unrelated waveforms (e.g., during a break in a task), and prevent over-clustering with the aid of metrics for comparing response profiles. Thus, the presented pipeline addresses the three-pronged objectives of algorithm-adaptability, rigorous validation, and human single-neuron recording optimization to support clinical and cognitive neuroscience applications.