Deep Learning Classification of Consciousness-Impairing Seizures with Intracranial Electroencephalography Recordings in Focal Epilepsy

Saturday, May 4, 2024

3:51 PM – 3:53 PM CT

Location: S405b

Presenting Author(s)

Derek J. Doss, BE

Medical Student
Vanderbilt University
Nashville, Tennessee, United States

Co-Author(s)

RR

Rohan Rashingkar, n/a

Undergraduate Student
Vanderbilt University, United States

Co-Author(s)

CB

Camden Bibro, BA

Research Assistant
Vanderbilt University, United States
Sarah K. Bick, MD (she/her/hers)

Assistant Professor of Neurological Surgery
Vanderbilt University Medical Center
Nashville, Tennessee, United States
CC

Catie Chang, PhD

Professor
Vanderbilt University, United States
Dario J. Englot, MD, PhD

Associate Professor of Neurological Surgery
Vanderbilt University Medical Center
Nashville, Tennessee, United States
KH

Kevin Haas, MD, PhD

Professor of Neurology
Vanderbilt University Medical Center, United States
Graham W. Johnson, PhD

Medical Student
Vanderbilt University
Antioch, TN, United States
VM

Victoria Morgan, Vanderbilt University

Professor
PhD, United States
Danika L. Paulo, MD

Resident
Vanderbilt University Medical Center
Nashville, Tennessee, United States
JS

Jared Shless, n/a

Medical Student
Harvard Medical School
San Francisco, California, United States
SW

Shawniqua W. Williams Roberson, MD

Fellow
Vanderbilt University, United States

Disclosure(s):

Derek J. Doss, BE: No financial relationships to disclose

Sarah K. Bick, MD: Varian Medical: Consultant (Terminated, February 28, 2023)

Dario J. Englot, MD, PhD: No financial relationships to disclose

Introduction: Epilepsy affects 1% of the global population and can consist of consciousness-impairing and consciousness-sparing seizures. Consciousness-impairing seizures can be particularly devastating for patients. Surgical treatment with closed loop neuromodulation may prevent loss of consciousness in these seizures, but electrographic differentiation of consciousness-impairing and consciousness-sparing seizures is challenging. Therefore, we sought to develop a deep learning network to detect consciousness-impairing seizures in focal epilepsy.

Methods: Stereotactic electroencephalography recordings were obtained for 73 patients in the Vanderbilt University Epilepsy Monitoring Unit. Each seizure was classified as consciousness-impairing (N=247) or consciousness-sparing (N=161). 10 seconds of preictal and 10 seconds of ictal data were collected for each seizure. The dataset was split into 5 folds and then augmented with random noise, frequency shifts, and DC shifts. A convolutional deep learning neural network with both temporal and spatial convolutions was developed. Combinations of preictal and ictal data from the entire 20-second epoch were evaluated with 10-second windows to determine the ideal combination to classify consciousness-impairing seizures. The model was evaluated on the completely held-out and unaugmented test dataset.

Results: The deep learning model accurately classified consciousness-impairing seizures with electrographic data alone. Furthermore, the best-performing models use a combination of both preictal, and ictal data and the worst performing models used only preictal or ictal data. The best performing model was sensitive to consciousness-impairing seizures, classifying them with 79.3% accuracy, 69.7% specificity, and 85.7% sensitivity.

Conclusion : Consciousness-impairing seizures in focal epilepsy can greatly impact quality of life. It may be possible to prevent the loss of consciousness with closed-loop surgical neuromodulation, but consciousness-impairing seizures are difficult to differentiate electrographically. Overall, this work demonstrates that it is possible to detect consciousness-impairing seizures in focal epilepsy and may lay the foundation for custom neuromodulation to preserve consciousness.