After clicking “Watch Now” you will be prompted to login or join.
Deep, Self-Supervised Learning for Patient-Specific Anomaly Detection in Stereoelectroencephalography
Anthony Costa, Icahn School of Medicine at Mount Sinai | Eric Oermann, Icahn School of Medicine at Mount Sinai
GTC 2020
We'll discuss methods for accurate, real-time detection of anomalous events in medical time-series data, with particular application to stereoelectroencephalographic (SEEG) interpretation and event localization for patients with epilepsy. Previous deep learning-based approaches to detecting EEG anomalies have been plagued by high false-positive rates and inability to detect anomalous events that fall below a statically-set threshold. You'll learn the benefits of a nonparametric approach that utilizes a dynamic thresholding method for event detection, producing significant performance improvements. We'll discuss basic approaches and challenges in anomaly detection in medical time series data, limitations encountered by previous deep-learning approaches to anomaly detection, and how to apply a nonparametric dynamic thresholding procedure to improve performance and mitigate false positive results.