Authors :
Presenting Author: Clarissa Hoffman, – UT Health Science Center - Houston
Yash Vakilna, MS – UT Health Science Center - Houston; Blanca Talavera, MD – UT Health Science Center - Houston; Norma Hupp, AS – UT Health Science Center - Houston; Johnson Hampson, MSBME – UT Health Science Center - Houston; Sandhya Rani, PhD – UT Health Science Center - Houston; Jaison Hampson, MBBS – UT Health Science Center - Houston; Samden Lhatoo, MD, FRCP – UT Health Science Center - Houston; Yuri Dabaghian, PhD – UT Health Science Center - Houston
Rationale:
Characterized by recurrent seizures, epilepsy is an incurable, chronic neurological condition that affects 50 million people worldwide. Despite the recurrent nature of epilepsy, the ability to predict seizure occurrence and its spread across the cortex remains elusive to date. We propose an approach for seizure detection that queries pattern dynamics in EEG and identifies atypical epochs at the level of patterns.
Methods:
We use three mathematical techniques to capture the dynamics of the pattern of peaks of the brainwaves. The first compares patterns of each brainwave to their statistically expected behavior and determines times and channels in which there is deviation from the trend. The second establishes how orderly the pattern is, for instance, whether patterns are statistically mundane or atypically periodic/clustered. Finally, the third method was used to identify statistical differences between patterns of signals recorded on different channels to visualize network level temporal changes before, during, and after seizures and gain a deeper understanding of seizure spread through the cortex.
Results:
We analyzed scalp EEG from seventeen patients and found substantial deviations of brain wave patterns during epileptic events followed by a prolonged period of sustained atypicality in the post ictal period. In fact, we observed statistically significant deviation from the trend in most channels up to 30 minutes before the seizure which could potentially be used clinically to herald an upcoming seizure. When comparing two channels, we observe periods of synchrony and asynchrony between both brain regions that correlate with both movement and changes in the patients’ sleep versus wake states. Seizures are marked by very rapid changes in the patterning of the brain rhythms. Interestingly, there is a prolonged period of time post-seizure characterized by atypical patterns. Furthermore, we discovered our technique is also highly sensitive to changes in the patients’ state, such as sleep vs awake.
Conclusions:
Ultimately, this work offers a novel semantics that allows analyzing epilepsy at the level of waveforms, using two independent, complementary techniques, and putting the results into a universal statistical perspective. It becomes possible to capture disturbances of the EEG activity in an intuitively transparent form, which brings together math and medicine to potentially advance clinical practice by facilitating the early detection of seizures.
Funding:
NIH R01NS110806
NSF 1901338
NIH R01AG074226