ICTALS Highlights 2022

Read below to learn more about the projects that were presented at this year’s ICTALS conference by members of the Davis Lab!


Quantifying Seizure Spread Using Deep Learning Algorithms to Localize Seizure Onset

Andrew Y. Revell, Erin C. Conrad, Brittany Scheid, Brian Litt, Kathryn A. Davis

  • Question: How can we best quantify seizure spread captured on EEG?
  • Methods: Researchers trained three deep learning algorithms on intracranial EEG data from 13 patients to classify interictal and ictal states on each channel.
  • Results: The deep learning models, along with line length, were able to differentiate interictal and ictal states in five test-set patients with varying levels of robustness.
  • Conclusion: Deep learning models can be used to quantify seizure spread and may localize seizure onset better than simple features when combined with predictive network models of spread.


Intracranial EEG structure-function coupling predicts surgical outcomes in focal epilepsy

Nishant Sinha, John S. Duncan, Beate Diehl, Fahmida A. Chowdhury, Jane de Tisi, Anna Miserocchi, Andrew W. McEvoy, Kathryn A. Davis, Sjoerd B. Vos, Gavin P. Winston, Yujiang Wang, Peter N. Taylor

  • Hypothesis: In epilepsy, surgery alters the structural brain network to control seizure activity emerging at the functional level. Given that surgery is a structural modification aiming to alter the function, researchers hypothesized that strong structure-function coupling improves post-operatively seizure control.
  • Methods: Researchers constructed structural and functional brain networks in 39 patients with medication-resistant focal epilepsy using multimodal data from intracranial EEG recordings, structural MRI (pre and post-surgery), and diffusion-weighted MRI (pre-surgery).
  • Results: At a whole network level, seizure-free patients had stronger structure-function coupling than not seizure-free patients regardless of the choice of interictal segment or frequency band. Structure-function coupling measures had the highest feature importance relative to clinical attributes, and they predicted seizure outcomes with an accuracy of 85% and sensitivity of 87%.
  • Conclusion: The underlying assumption that the structural changes induced by surgery would translate to the functional level to control seizures is valid when the structure-functional coupling is strong.


Sleep and Epilepsy

Erin C. Conrad, Andrew Y. Revell, James J. Gugger, Russell T. Shinohara, Brian Litt, Eric D. Marsh, Kathryn A. Davis

  • Question: What is the effect of sleep and seizures on spikes, and do sleep and seizure-related changes in spikes localize the seizure onset zone?
  • Methods: Researchers performed a retrospective analysis of intracranial EEG data from 96 patients with drug-resistant focal epilepsy.
  • Results: The alpha-delta power ratio accurately classified wake from sleep periods (AUC = 0.90). A machine-learning classifier incorporating only spike rates and sleep/wake state accurately identified the seizure onset zone (AUC = 0.78).
  • Conclusion: The change in spike rates surrounding seizures can be used to localize temporal versus extratemporal lobe epilepsy. Spikes are more frequent and better localize the seizure onset zone in sleep.


Epilepsy Surgery

John M. Bernabei, Nishant Sinha, T. Campbell Arnold, Erin Conrad, Ian Ong, Akash R. Pattnaik, Joel M. Stein, Russell T. Shinohara, Timothy H. Lucas, Dani S. Bassett, Kathryn A. Davis, Brian Litt

  • Hypothesis: Researchers hypothesize that using a normative iEEG atlas to benchmark deviations from normal brain dynamics provides a data-driven method to identify surgical targets.
  • Methods: Researchers constructed a normative iEEG atlas by augmenting a 106-subject normative iEEG atlas from the Montreal Neurological Institute (MNI) with 60 subjects carefully selected from the Hospital of Pennsylvania (HUP). Researchers quantitatively compared epileptic iEEG channels to normative data and mapped patient-specific abnormalities.
  • Results: The results demonstrate that for seizure onset zones (SOZ) within the mesial temporal lobe, measures of connectivity abnormality provide greater distinguishing value than univariate measures of abnormal neural activity. Furthermore, patients with longer diagnoses of epilepsy have greater abnormalities in connectivity.
  • Conclusion: This study establishes a data-driven method to guide epilepsy surgery by aggregating iEEG studies.


Read more about these projects HERE.

Penn Presents | Davis Lab Summer Work 2022

As a Research Assistant to Dr. Davis at Penn’s Center for Neuroengineering and Therapeutics, Hannah Gonzalez worked alongside Dr. Arnold to train a deep learning model to automate resection cavity segmentation on postoperative MRI of epilepsy patients to help physicians quantify removed brain structures.

The purpose of the project was to expand on a previous paper about Deep Learning-Based Automated Segmentation of Resection Cavities on Postsurgical Epilepsy MRI by adding a step in the data preprocessing and training of the model. After training 3 different models and running inference and the majority vote algorithm, it was determined that the Axial/Coronal model is the highest performer.

Future work in investigating the sagittal model to see how its performance can be improved.

Read more about this project HERE.