|Video of the Researcher|
Dr. Gary Hasey
Using machine learning analysis of EEG data to establish psychiatric diagnosis and to determine optimal treatment of major depressive disorder
|Type of researcher|
|Introduce yourself, your experience and your credentials||
Using machine learning analysis of EEG data to establish psychiatric diagnosis and to determine optimal treatment. Transcranial magnetic stimulation as a treatment for major depressive disorder.
|Describe your research||
The problems that we have struggled to address relate to the imprecision of Psychiatry by imprecision I mean much of what I do as a psychiatrist is impressionistic guesswork. Diagnosis doesn’t have any blood test or any brain image that will tell us undisputably what’s wrong with a given person it’s all based on impressions and history.
Similarly it’s very difficult to determine the correct treatment careful study of this issue in the area of mood disorders has shown us that even the best psychiatrists choose the correct treatment for a given patient only about 37% of the time. So what we have been doing over the last decade or so is using brain electrical activity patterns as a source for our decision making what we are basically doing is compressing years or months well probably years maybe even decades of work into a few days of processing and mathematical analysis.
So we have been quite happy with the results our algorithms have been capable of being trained to diagnose patients with a high degree of precision. So our algorithms can diagnose very difficult to detect illnesses like bipolar disorder with about 80% plus accuracy.
Furthermore the analyses can determine within each diagnostic cluster in this multi-dimensional EEG world, it can detect sub clusters that correlate with response to a particular biological treatment so we can predict response to a particular drug, a particular type of psychotherapy, or particular brain stimulation technology such as repetitive transcranial magnetic stimulation.
|Explain its significance||
So this is a huge advance. Clinically as I said to begin with my science is imprecise we are impressionistic and treatment devolves to what might be charitably called serial trial and error. So now instead of that trial and error process we have a very precise quantitative way to determine what the best treatment for a given individual would be.
We have also done some work with other data sets, for example in the area of post-traumatic stress disorder, suicide among veterans and serving military personnel is a very major problem. So we’ve been able to identify using machine learning a subset of 25 separate items or even a smaller subset of ten items that allow us to determine if a given service man or woman has contemplated taking their own life through suicide, with about 80 plus percent accuracy.
So this is a way to access these dangerous kinds of thoughts without asking that very difficult question and this was only possible using machine learning. So in a nutshell that’s the kind of work my group has been doing for the last decade. We’ve predicted response to drugs, to cognitive behavior therapy, to electroconvulsive therapy, to transcranial magnetic stimulation, and to an anti-psychotic drug called clozapine. So we now have a quantitative way to do my job better.
McMaster University, Main Street West, Hamilton, ON, Canada
|Type of institution|