Dawant, Benoît M.

Abstract How do brain networks limit seizure activity? In the interictal suppression hypothesis, we recently postulated that high inward connectivity to seizure onset zones (SOZs) from non-involved zones (NIZs) is a sign of broader network suppression. If broad networks appear to be responsible for interictal SOZ suppression, what changes during seizure initiation, spread and termination? For patients with drug-resistant epilepsy, intracranial monitoring offers a view into the electrographic networks that organize around and in response to the SOZ. In this manuscript, we investigate network dynamics in the peri-ictal periods to assess possible mechanisms of seizure suppression and the consequences of this suppression being overwhelmed. Peri-ictal network dynamics were derived from stereo electroencephalography recordings from 75 patients with drug-resistant epilepsy undergoing presurgical evaluation at Vanderbilt University Medical Center. We computed directed connectivity from 5 s windows in the periods between, immediately before, during and after 668 seizures. We aligned connectivity matrices across seizures and patients, then calculated net connectivity changes from the SOZ, propagative zone and NIZ. Across all seizure types, we observed two phases: a rapid increase in directed communication towards the SOZ, followed by a collapse in network connectivity. During the first phase, SOZs could be distinguished from all other regions (one-way ANOVA, P = 8.32 × 10−19–2.22 × 10−7). In the second phase and post-ictal period, SOZ inward connectivity decreased yet remained distinct (one-way ANOVA, P = 2.58 × 10−10–1.66 × 10−2). NIZs appeared to drive increased SOZ connectivity while intra-NIZ connectivity decreased concordantly. Stratifying by seizure subtype, we found that consciousness-impairing seizures decrease inward connectivity from the NIZ earlier than consciousness-sparing seizures (one-way ANOVA, P < 0.01 after false discovery correction). Tracking network reorganization against a surrogate for seizure involvement highlighted a possible antagonism between seizure propagation and the ability of the NIZ to maintain high connectivity to the SOZ. Finally, we found that inclusion of peri-ictal connectivity improved the accuracy of SOZ classification from previous models to a combined area under the curve of 93%. Overall, NIZs appear actively to increase inhibitory signaling towards the SOZ during seizure onset, possibly to thwart seizure activity. Although inhibition appears insufficient to prevent seizure onset, the inability to restrict seizure propagation might contribute to loss of consciousness during larger seizures. Dynamic connectivity patterns uncovered in this work might: (i) facilitate accurate delineation of surgical targets in focal epilepsy; (ii) reveal why interictal suppression of SOZs might be insufficient to prevent all seizures; and (iii) provide insight into mechanisms of loss of consciousness during certain seizures.

Objectives/Goals: Manual skin assessment in chronic graft-versus-host disease (cGVHD) can be time consuming and inconsistent (>20% affected area) even for experts. Building on previous work we explore methods to use unmarked photos to train artificial intelligence (AI) models, aiming to improve performance by expanding and diversifying the training data without additional burden on experts. Methods/Study Population: Common to many medical imaging projects, we have a small number of expert-marked patient photos (N = 36, n = 360), and many unmarked photos (N = 337, n = 25,842). Dark skin (Fitzpatrick type 4+) is underrepresented in both sets; 11% of patients in the marked set and 9% in the unmarked set. In addition, a set of 20 expert-marked photos from 20 patients were withheld from training to assess model performance, with 20% dark skin type. Our gold standard markings were manual contours around affected skin by a trained expert. Three AI training methods were tested. Our established baseline uses only the small number of marked photos (supervised method). The semi-supervised method uses a mix of marked and unmarked photos with human feedback. The self-supervised method uses only unmarked photos without any human feedback. Results/Anticipated Results: We evaluated performance by comparing predicted skin areas with expert markings. The error was given by the absolute difference between the percentage areas marked by the AI model and expert, where lower is better. Across all test patients, the median error was 19% (interquartile range 6 – 34) for the supervised method and 10% (5 – 23) for the semi-supervised method, which incorporated unmarked photos from 83 patients. On dark skin types, the median error was 36% (18 – 62) for supervised and 28% (14 – 52) for semi-supervised, compared to a median error on light skin of 18% (5 – 26) for supervised and 7% (4 – 17) for semi-supervised. Self-supervised, using all 337 unmarked patients, is expected to further improve performance and consistency due to increased data diversity. Full results will be presented at the meeting. Discussion/Significance of Impact: By automating skin assessment for cGVHD, AI could improve accuracy and consistency compared to manual methods. If translated to clinical use, this would ease clinical burden and scale to large patient cohorts. Future work will focus on ensuring equitable performance across all skin types, providing fair and accurate assessments for every patient.