Authors :
Presenting Author: Hana Farzaneh, MD – Boston Children's Hospital
Alyssa Ailion, PhD – Boston Children's Hospital & Harvard Medical School
Rebecah Kaplun, BS – Boston Children's Hospital
Dana Martino, BS – Boston Children's Hospital
Rationale:
Accurate delineation of the epileptogenic network in pediatric drug-resistant epilepsy (DRE) is essential, yet resource-intensive. We report the integration of a novel automated neuroinformatics platform into our surgical workflow, streamlining multimodal image fusion, sEEG contact localization, ictal-clip extraction, and network analysis. We finalized the retrospective performance of a seizure-free cohort and set the stage for an ongoing prospective evaluation.
Methods:
Twenty consecutive children who achieved Engel I outcomes at ≥ 12 months postoperatively were analyzed. Data uploaded to the platform included preoperative MRI, post-implant CT, postoperative MRI, and full-length sEEG recordings with marked ictal onset.
Pipeline (fully executed on the FIND Neuro platform):
- Image processing: automated brain extraction and cortical reconstruction; rigid CT: MRI registration.
- Contact localization: A CNN-based neural network segments the sEEG contacts and a post-processing pipeline refines their positions. Once the detector proposes the contact coordinates, clinicians can interactively adjust each point in the 3D viewer.
- Signal preparation: adaptive clipping (60s per ictal event) and resampling to 512 Hz.
- Network inference – Time-varying vector-autoregressive Granger causality, with lags chosen by likelihood-ratio testing, constructs a directed acyclic seizure-propagation graph for each clip; the sum of a node’s outgoing causal weights (out-degree) quantifies its “criticality.”
- 3-D visualization: MRI, reconstructed skull, and color-coded contacts rendered in an interactive WebGL viewer.
Endpoints: (i) localization error vs. CT ground truth, (ii) overlap of high-criticality contacts with treated contacts, and (iii) workflow time compared with the previous semi-manual pipeline.Results:
- Localization accuracy: All contacts were accurately identified after minimal manual adjustment
- Concordance: The top five high-criticality contacts were within the surgical treatment volume.
- Efficiency: Total processing time < 30 minutes after ~3 hours of MRI reconstruction
- Usability: All cases were integrated into the platform and could be used in the surgical planning pipeline.
Conclusions:
With FIND Neuro’s funded integration, we achieved sub-2 mm contact localization and rapid epileptogenic network reconstruction while maintaining strong concordance with successful surgical targets. These results justify the prospective study now underway, which will quantify the platform’s impact on real-time surgical planning in newly implanted patients.
Funding: Industry-funded project (FIND Neuro)