Deep brain local field potential neural decoders for real-time computer interfacing
File(s)
Author(s)
Martineau, Thomas
Type
Thesis
Abstract
The progressive adoption of deep brain stimulation (DBS) therapy to treat neurological disorders offers a unique opportunity to research a new class of invasive brain-computer interface (BCI). DBS electrodes can record electrophysiological activity or local-field potentials (LFP) within the subcortical structures, particularly the subthalamic nucleus (STN). Although STN-LFP biomarkers have been correlated to multiple motor processes, including motion, force generation, action sequencing, and gait, subcortical structures have not received nearly as much attention as other motor cortical areas. Instantaneous STN interfacing could inform the control of a wide range of medical devices and possibly neuroprosthetics.
This work provides evidence that STN-LFP decoders can predict volitional motor states and presents solutions tested on patients' recordings undergoing DBS surgery. Decoding needed to be further validated under real-time constraints. Part of this work was dedicated to implementing, testing, and comparing compatible algorithms, particularly for improved time-frequency feature extraction across multiple signal channels. Decoders were customised for different decoding tasks, and parameter tuning methods were also introduced. Subsequent experiment results showed that these strategies helped increase performance.
The first experiment decoded discrete motor states from STN-LFP to achieve action detection and left vs right-hand selection. Results showed that those motor state variables could be predicted under asynchronous and real-time constraints (median 0.7-0.3 action detection rate and 0.7 left vs right classification accuracy). However, performance was very variable across study participants. Evidence pointed to an unequal signal-to-noise ratio of STN-LFP biomarkers in different recordings, which correlated directly to individual performance ceilings.
The second set of BCI tasks focused on continuous state decoding for fine motor control. New force tracking paradigms revealed new association of biomarkers with motor states of different dynamical orders. Some decoding algorithms showed favourable results.
This work concludes that STN-LFP neural decoders could have promising applications, given some adaptations and further innovation.
This work provides evidence that STN-LFP decoders can predict volitional motor states and presents solutions tested on patients' recordings undergoing DBS surgery. Decoding needed to be further validated under real-time constraints. Part of this work was dedicated to implementing, testing, and comparing compatible algorithms, particularly for improved time-frequency feature extraction across multiple signal channels. Decoders were customised for different decoding tasks, and parameter tuning methods were also introduced. Subsequent experiment results showed that these strategies helped increase performance.
The first experiment decoded discrete motor states from STN-LFP to achieve action detection and left vs right-hand selection. Results showed that those motor state variables could be predicted under asynchronous and real-time constraints (median 0.7-0.3 action detection rate and 0.7 left vs right classification accuracy). However, performance was very variable across study participants. Evidence pointed to an unequal signal-to-noise ratio of STN-LFP biomarkers in different recordings, which correlated directly to individual performance ceilings.
The second set of BCI tasks focused on continuous state decoding for fine motor control. New force tracking paradigms revealed new association of biomarkers with motor states of different dynamical orders. Some decoding algorithms showed favourable results.
This work concludes that STN-LFP neural decoders could have promising applications, given some adaptations and further innovation.
Version
Open Access
Date Issued
2022-04-16
Date Awarded
01/07/2023
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Vaidyanathan, Ravi
Sponsor
Engineering and Physical Sciences Research Council
Grant Number
EP/L016737/1
Publisher Department
Mechanical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)