ABSTRACT
An EEG measures electrical activity of the brain from the scalp surface and is the primary clinical tool for diagnosing epilepsy and strokes. Clinical use of EEGs is rapidly increasing as new applications are being developed, including the diagnosis of sleep disorders, head-related trauma injuries, and Alzheimer’s. Furthermore, with the advent of wireless technology, long-term monitoring occurring over a period of time from several hours to several days has become possible, overwhelming clinicians and rapidly outstripping the resources available to manually interpret such data. The development of a system that can automatically interpret an EEG allows healthcare providers to keep pace with the growing demand for this diagnostic tool and would provide real-time alerting of potentially life-threatening conditions. Machine learning has made tremendous progress over the past three decades in many fields due to rapid advances in low-cost highly parallel computational infrastructure, powerful machine learning algorithms, and, most importantly, big data. In this chapter, we describe the steps involved in the development and evaluation of a high-performance machine learning system that automatically identifies key events in an EEG signal at performance levels close to human performance. We describe a unique open source big data resource known as the TUH EEG Corpus that enabled the development of this technology by supporting the application of state-of-the-art statistical modeling. We also describe several related applications including the detection of abnormal EEGs and seizures. The underlying technology common to these systems is based on a combination of hidden Markov models and deep learning.
