Michael Caiola , Avaneesh Babu, Meijun Ye
Abstract
Traumatic Brain Injury (TBI) and stroke are devastating neurological conditions that affect hundreds of people daily. Unfortunately, detecting TBI and stroke without specific imaging techniques or access to a hospital often proves difficult. Our prior research used machine learning on electroencephalograms (EEGs) to select important features and to classify between normal, TBI, and stroke on an independent dataset from a public repository with an accuracy of 0.71. In this study, we expanded to explore whether featureless and deep learning models can provide better performance in distinguishing between TBI, stroke and normal EEGs by including more comprehensive data extraction tools to drastically increase the size of the training dataset. We compared the performance of models built upon selected features with Linear Discriminative Analysis and ReliefF with several featureless deep learning models.
Introduction
Traumatic brain injury (TBI) and stroke directly affect millions of people annually [1, 2]. In fact, the Centers for Disease Control and Prevention (CDC) estimates 176 Americans die from TBI-related injuries each day [3] and an American suffers a stroke every 40 seconds [2]. Treatment and recover options differ depending on the severity of TBI or stroke, however, without proper diagnosis or immediate detection, many cases go untreated.
Materials and method
All patient data in the database were de-identified. Therefore, this study did not constitute human subjects research, and was exempted from Food and Drug Administration institutional review board review.
Results
A set of 1406 features were calculated from each EEG and two methods of feature selection were used to reduce the most important features: LDA and ReliefF. LDA was calculated using the full set of 1406 features and the entire set of chosen segments. Overall, LDA selected 192 features with the highest percentage of the features chosen in the delta, theta, alpha, mu, beta, and gamma relative PSDs, entropy, standard deviation, and delta absolute PSD (Fig 6a).
Discussion
TUEG is a large public database that has been a common resource for those studying neurological diseases (e.g., TBI, seizures, and epilepsy) [22, 26] as well as other data mining initiatives [38]. The ability to extract information from large datasets is useful but also presents challenges in terms of selectivity. For example, since TUEG is not a disease-specific database, we only found ∼ 14% of the data useful for our application.
Conclusion
There is great interest in the academic community and by clinicians and device manufacturers in developing biomarkers and diagnostic tools for neurological disease. Our goal in this project was to develop tools, models, and biomarkers to enhance the utility of EEG in TBI and stroke diagnoses.
Acknowledgments
The authors would like to thank Drs. Ramin Bighamian, Daniel X. Hammer, and Edward Nyman for reviewing and editing the paper.
Citation: Caiola M, Babu A, Ye M (2023) EEG classification of traumatic brain injury and stroke from a nonspecific population using neural networks. PLOS Digit Health 2(7): e0000282. https://doi.org/10.1371/journal.pdig.0000282
Editor: Henry Horng-Shing Lu, National Yang Ming Chiao Tung University, TAIWAN
Received: March 9, 2023; Accepted: May 30, 2023; Published: July 6, 2023
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: Data used in this study was obtained from the publicly available Temple University Hospital EEG Corpus (https://isip.piconepress.com/projects/tuh_eeg/). All machine learning and deep learning models, as well as feature selection results and tools will be available on the FDA / CDRH / OSEL / Division of Biomedical Physics GitHub (https://github.com/dbp-osel/).
Funding: This study was supported by internal funding from Division of Biomedical Physics, Office of Science and Engineering Laboratories, Center for Devices and Radiological Health of the U.S. Food and Drug Administration. The funders had no additional role in study design, data collection and analysis, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Source: https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000282#ack
