Authors :
Presenting Author: Luis Sanchez, PhD Student – Johns Hopkins University
Amir Daraie, PhD Student – Johns Hopkins University; Patrick Myers, PhD Student – Johns Hopkins University; Kristin Gunnarsdottir, PhD – Johns Hopkins University; Jorge Gonzalez-Martinez, MD – University of Pittsburgh Medical Center; Joon Kang, MD – Johns Hopkins University School of Medicine; Sridevi Sarma, PhD – Johns Hopkins University
Rationale:
Epilepsy is a neurological condition characterized by recurrent abnormal electrical activity in the brain or seizures. Surgical intervention is a potential solution for 30% of the epilepsy patients who do not respond to anti-epileptic drugs. However, the success rates vary (30% to 70%), and there is a lack of reliable indicators for identifying the epileptogenic zone (EZ). Additionally, there is currently no tool available to assess the likelihood of surgical success.
Methods:
In this research study, we constructed personalized dynamic "brain" network models (DNMs) based on minutes of inter-ictal (between-seizure) intracranial EEG (iEEG) data from each patient. Using these DNMs, we conducted virtual resections by targeting the clinically resected brain regions for each patient. In silico, we simulated the removal of the EZ by neutralizing its signals, retrained the DNM, and then analyzed the changes in network properties after the simulated surgery. Specifically, we examined a novel iEEG biomarker known as "source-sink," which characterizes the network properties of the EZ in the DNM before and after the virtual surgery. Sources refer to brain regions that exert a significant influence on other regions but are not influenced themselves by the network, while sinks represent regions that are heavily influenced by other regions but do not exert influence on others (Gunnarsdottir 2022). Based on this, we hypothesized that successful surgeries would be predicted by the reduction of both strong sinks and sources that were present in the epileptic brain prior to the surgery in the virtual post-surgery scenario.
Results:
We tested our hypothesis on a cohort of 48 patients who underwent surgical treatment at three different centers: Cleveland Clinic, NIH, and Johns Hopkins. Figure 1 illustrates the distribution of sources and sinks in a successful surgical case (Engle Score 1). Notably, when the clinically identified EZ channels were virtually removed, both the sources (depicted in blue) and sinks (depicted in red) exhibited a significant reduction (lower panel, left to right). Furthermore, the count distribution of sources and sinks assumed a normal distribution shape (upper panel, left to right). Prior to the surgery, the EZ functioned as strong sinks influenced by strong sources, whereas post-surgery, no strong sources or sinks were observed. To assess the similarity of the source-sink distribution to a normal distribution, we employed the Jarque-Bera (JB) normality test, which measured the skewness and kurtosis. Our findings demonstrated that the JB test statistic for successful cases shifted from large values pre-surgery to values close to zero post-surgery, with p-values well above 0.05. Figure 2 presents boxplots that depict the difference in p-values between pre- and post-surgery for patients categorized by their Engle score.
Conclusions:
These findings suggest that DNMs combined with this biomarker may be used to test surgical plans prior to surgery and may improve outcomes.
K. M. Gunnarsdottir et al., “Source-sink connectivity: a novel interictal EEG marker for seizure localization,” Brain, vol. 145, no. 11, pp. 3901–3915, Nov. 2022, doi: 10.1093/brain/awac300.
Funding: none