Jim Reilly

Professor Emeritus, Electrical and Computer Engineering

0000-0002-4109-1186

One of Prof. Reilly’s research interests is machine learning analysis of the electroencephalogram (EEG) to help understand the human brain. For example, the best method that psychiatrists currently have available for treating severe depression is trial and error; i.e., anti-depressant medications are tried on an iterative basis until one is found that works. This process can take many months before the patient experiences relief, during which time they suffer badly and are at an elevated risk of suicide. We have developed a machine learning approach for analyzing the pre-treatment EEG to determine the most effective treatment for a specific individual, at the outset of treatment. Thus, with our method, the chances of the psychiatrist prescribing an effective medication on the first iteration increases to about 85%, compared to the current value of about 33%. This drastically reduces the time to remission and alleviates burden on the health care system.

A second example of our research is predicting whether or not a coma patient will emerge. The mismatched negativity (MMN) component of the EEG is the response of the subject when a deviant tone stimulus is interspersed within a series of standard tones. It has been well established that the presence of the MMN in a coma patient’s EEG responses is indicative of emergence. The difficulty however is that this component is very difficult to detect. With current methods, the MMN is not discernable approx. 70% of the time even when it is actually present. We have developed a machine learning approach for detecting the MMN component with about 92% accuracy. Therefore, with this new approach, we can determine with confidence whether the patient will or will not emerge.