Abstracts

Rationale: More than ever before, breakthroughs in neuroscience are generated by multi-center collaborations. Central to these collaborations is the ability to rapidly collect, annotate and analyze shared data across geographically dispersed centers. We present a new, cloud based, platform for this purpose, centered on epilepsy and funded by NINDS. The platform provides a web-based viewer for data and results, computational power for parallel analysis, a repository for sharing algorithms and their source code, and communication between collaborating centers. It allows the user to analyze the data by providing direct access to the data using commonly used software development platforms such as Matlab, and provides the capability to run analysis in parallel on multiple worker nodes to maximize efficiency and speed. Methods: In a collaborative effort to demonstrate the effectiveness and features of the international EEG portal, data from a single patient was analyzed by 6 independent epilepsy research centers across the world. The analyzed dataset contained multichannel data belonging to a patient with neocortical epilepsy implanted with macro- and micro-wire electrode arrays, which were sampled at 5000Hz (0.5-1000 Hz band-pass filtered). Investigators were free to use any methods they would like and were asked to upload their results to the database. The interaction with the portal was optimized to impose minimal changes to code the user might already be using locally. Four centers provided a high frequency oscillations (HFO) detector; one center developed a spike detector and one center provided a seizure predictor. The entire experiment was conducted over 2 weeks. Results: HFO detections correlated with the vicinity of the clinician-marked seizure onset zone, though there was great variability in the number of HFOs detected across different algorithms. This highlights the need for a centralized EEG platform where results from analysis can be compared on standardized datasets. Further investigation will allow quantitative comparisons between the qualities of the provided algorithms. Localization of interictal spikes did not correlate well with seizure onset zone and did not correlate well with the HFO events. Conclusions: Sharing data and validating analysis using a standardized dataset requires significant efforts from the epilepsy research community, but has potential to greatly accelerate collaborative research. This is not uncommon in other research fields. Our results clearly identify the need for shared data sets that can validate and compare algorithms from different research centers. With this platform in place, collaborative research using these tools can be rapidly accelerated. Challenges will include aligning incentives for investigators to share data, algorithms and knowledge, as current systems for funding and promotion can discourage these endeavors.