Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Engineering Science

Program

Biomedical Engineering

Collaborative Specialization

Machine Learning in Health and Biomedical Sciences

Supervisor

Khan, Ali R.

2nd Supervisor

Lau, Jonathan C.

Co-Supervisor

Abstract

This thesis serves to address the problem of non-standardized preprocessing of intracranial electroencephalography (iEEG) recordings by implementing a software workflow that compiles some of the most common steps followed for the preparation of this type of data. This workflow improves the consistency, replicability, and ease of use of iEEG preprocessing, facilitating the replication and extension of previous studies and the combination of separately preprocessed inter-institutional datasets. Automatic detection of artifacts for iEEG data was also explored as a potential step to include in the preprocessing workflow. Despite training the models with cross-institutional data, poor performance was observed when tested on external datasets, showing the need for more and higher quality cross-institutional datasets to develop truly generalizable models. Future work is needed to include other common preprocessing methods, validate the developed tool with external datasets, and ensure compliance with BIDS standards to establish a standardized preprocessing tool for iEEG recordings.

Summary for Lay Audience

In this thesis, I seek to address a common problem in neuroscience research related to a brain monitoring technology called intracranial electroencephalography (iEEG). This technology, where electrodes are surgically implanted inside of the skull for recordings, is essential for understanding, treating and completing research on epilepsy. Currently, there is no standardized preparation of iEEG data for epilepsy research, which hinders the comparison between studies developed in different institutions and extension of previous findings.

To tackle this issue, I developed a new software workflow that compiles common preprocessing steps for iEEG data. This tool aims for a more consistent preparation of the data between different research settings, making it easier for researchers to replicate and extend findings and to combine data from different institutions. This initial version of the workflow only supports preprocessing of stereoelectroencephalography (SEEG) data, a type of iEEG, chosen due to its common usage in clinical settings compared to other iEEG techniques.

The presented thesis presents a validation of the developed workflow using local datasets. More testing with external datasets is needed to confirm broader usability and generalizability. Additionally, the workflow currently supports only SEEG recordings and limited file formats and preprocessing methods, which require expansion to increase their utility.

I also explored automatic artifact detection for iEEG, which is another fundamental step in iEEG preprocessing with no clear standardized or validated methodology. Although the models tested in this work show promise, they did not perform well enough to be included in the final preprocessing workflow developed. This highlights the need for larger and higher-quality datasets from multiple institutions to develop reliable models that can be standardized in the preprocessing workflow.

Despite these advances, several primary limitations remain to ensure the workflow is universally applicable in epilepsy and neuroscience research. These include the need to perform more extensive testing, increase the number of available methods, improve computational efficiency, and adhere to the BIDS standards. This project not only provides a practical tool for researchers, but also summarizes the challenges and complexities involved in managing and analyzing iEEG data, highlighting the need for further innovation in this field.

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS