NYU Grossman School of Medicine works to predict sudden death by epilepsy

The research team at NYU Grossman School of Medicine has very capable on-premises servers using graphics processing units (GPUs) that provide important computing power for machine learning workloads. But those GPUs are shared by a number of researchers.

Given that a single deep learning simulation can take up to 24 hours, it can be a slow process for a researcher to gain access to one of these on-premises devices. In addition, getting administrational rights for these local computers can cause further delays.

NYU Grossman School of Medicine researchers wanted to expand their work to more projects, including studies that could run parallel to each other rather than waiting in queue for university computing resources. The team therefore needed access to more computing power to train the cutting-edge deep learning models faster.

A number of Oracle Cloud Infrastructure (OCI) features appealed to the researchers at NYU Grossman School of Medicine. They could access OCI GPUs instantly, which eliminated the wait time it took to access on-premises computers. Because OCI is remotely accessible, it was easier for the team to collaborate with external, offsite partners. Internal collaboration was also more straightforward, since all team members needed was a computer and an internet connection.

OCI also enabled the researchers to configure the environment to their specific needs, letting them create a custom sandbox cloud computing environment.

The team received an Oracle for Research Project Award, providing access to Oracle Cloud credits, computational resources, hands-on technology consultations, and assistance in promoting and publishing their work.

With OCI, the research team now has access to on-demand GPUs so it can do more research that applies deep learning and other machine learning techniques to the study of epilepsy and related conditions. That new research is underway and powered by OCI.

Using deep learning for this work shows considerable promise. In a multi-cohort study using university GPUs, the team at NYU Grossman School of Medicine compiled and analyzed electroencephalography (EEG) and electrocardiogram (ECG) data from patients recruited at eight epilepsy research centers in the US and Australia. Researchers then used machine learning to identify biomarkers that play a pivotal role in Sudden Unexpected Death in Epilepsy (SUDEP) occurrences. By determining some of the EEG patterns in the brain that preceed SUDEP, the team created a possible new approach to help identify who is at greatest risk for SUDEP.