We have recently developed flexible, active, multiplexed recording devices for high resolution interface with large, clinically relevant areas of the brain. While this technology has enabled a much finer view of the electrical activity of the brain, the analytical methods to process, categorize and respond to the huge volumes of data produced by these devices have not yet been developed. Many existing neurological data analyses rely on manual inspection.
One major application of our high resolution device is to record and examine brain signals in patients with epilepsy. Preliminary analyses of our high resolution data have discovered repetitive spatiotemporal epileptiform discharge (or spike) patterns that initiate and terminate seizures, not seen by standard electrodes. Understanding the ordering and relationships between these patterns is a key to develop better seizure detection and prediction techniques and ultimately better therapy. The goal of this research is to develop efficient and automated tools for spike segmentation, spike categorization and spike trajectory prediction, by combining exciting recent developments in video analysis and machine learning.
The research comprises four interconnected components. The first component will develop techniques for detecting and isolating spike segments, followed by extracting features that capture the spatial-temporal motion pattern of each spike. The second component will analyze distribution of spike patterns in different feature spaces (original as well as nonlinearly transformed), and identify a feature space where clusters with different motion patterns are separable. The third component will develop algorithms for seizure prediction based on not only the spatial temporal pattern of the current spike, but also the patterns of past spikes. Accurate seizure prediction will enable pre-emptive brain stimulation to prevent or suppress seizures. The fourth component will develop methods to predict spike wavefront locations, to enable real-time responsive stimulation at the anticipated wavefront location to suppress seizures.
This material is based upon work supported by the National Science Foundation with Grant Number 1422914.
Related Publications and technical reports
- Yilin Song, Yao Wang and Jonathan Viventi,Adversarial autoencoder analysis on human μECoG dataset.
- Yilin Song, Yao Wang and Jonathan Viventi, Multi Resolution LSTM For Long Term Prediction In Neural Activity Video.
- Yilin Song, Yao Wang and Jonathan Viventi, Unsupervised Learning of Spike Pattern for Seizure Detection and Wavefront Estimation of High Resolution Micro Electrocorticographic (μECoG) Data, under review for IEEE Trans. Nanobioscience.
- Yilin Song, Jonathan Viventi, and Yao Wang. "Diversity encouraged learning of unsupervised LSTM ensemble for neural activity video prediction."
- B. Akyildiz, Y. Song, J. Viventi, and Y. Wang, Improved Clustering of Spike Patterns Through Video Segmentation and Motion Analysis of Micro Electrocorticographic Data", IEEE Signal Processing in Medicine and Biology Symposium, 2013.
- Y. Song, B.Akyildiz, J.Viventi and Y.Wang, "Improved Clustering of Spike Patterns Through Video Segmentation and Motion Analysis of Micro Electrocorticographic Data", The 6th International Workshop on Seizure Prediction, 2013, San Diego [poster]
- Y.Song, J.Viventi and Y.Wang, "Seizure Prediction Through Manifold Clustering And Temporal Analysis of Micro Electrocorticographic Data", The Greater New York Area Multimedia and Vision Meeting, 2014, New York [poster]