Event detection and classification from multimodal time series with application to neural data.

Area of Science:

• Signal Processing
• Computational Neuroscience
• Data Science

Background:

• Event detection in time-series data is crucial but challenging with multimodal signals.
• Existing methods like matched filters are limited to single data types (e.g., Gaussian noise).
• Neuroscience experiments often generate multimodal data (e.g., local field potentials and neuronal spikes).

Purpose of the Study:

• To develop a novel algorithm for event detection in multimodal time-series data.
• To address the limitations of current methods in handling combined Gaussian and point-process signals.
• To enable simultaneous estimation of event times and classes from diverse data streams.

Main Methods:

• Developed the multimodal event detector (MED) algorithm.
• Formulated a multimodal likelihood function for Gaussian and point-process observations.
• Derived a maximum likelihood estimator for simultaneous event time and class estimation.
• Introduced a cross-modal scaling parameter to manage model mismatch.

Main Results:

• MED successfully detected event onset and classified event direction in simulated and real neural data (eye-movement task).
• The algorithm effectively integrated information across different data modalities.
• Multimodal performance of MED surpassed unimodal approaches.

Conclusions:

• The MED algorithm provides a robust solution for event detection in multimodal time-series data.
• This method has significant potential for discovering latent events in neural activity.
• MED can advance brain-computer interfaces for naturalistic applications.