BioSignal Analytics participates in the Phase 1 Ventures program at the University City Science Center, a startup accelerator for “long horizon” technologies in the healthcare, materials and energy industries. We have identified real-time seizure detection in an ICU environment as a first commercial application for AutoEEG. EEG is the primary tool for diagnosis of brain related illnesses, especially for patients with an altered mental status. Studies have shown changes in EEG can be correlated with seizures, ischemia and brain swelling. Additionally, continuous EEG monitoring of critical care patients with insults to the brain is important for directing medical interventions to help prevent secondary injury.
Analysis of EEG signals requires a highly trained neurologist, and is time consuming and expensive since identifying rare clinical events requires analysis of long data streams collected over several hours or days. In an ICU environment, facilities must arrange for 24/7 neurologist coverage to interpret results in a timely manner. High performance real-time detection of abnormal brain events will increase the use of continuous brain monitoring, by decreasing the cognitive burden placed on person responsible for monitoring results and helping to direct intervention.
AutoEEG is based on proven, advanced, deep learning technology. It will expedite making informed medical decisions that can potentially reduce complications and improve outcomes in brain injury and disease. This innovation is needed because existing approaches to seizure detection involve limited technologies that perform unacceptably in clinical settings. State of the art machine learning algorithms that employ high dimensional models have yet to be applied because data resources to support training of such systems are lacking. Our proposed approach directly addresses these issues through the use of (1) big data resources such as the Temple University Hospital (TUH) EEG Corpus, (2) state of the art machine learning algorithms based on advanced statistical models, and (3) a combination of spatial and temporal context, adaption and normalization algorithms to reduce false alarm rates.