- 222 views
- 562 downloads
Learning Models for Psychiatric Diagnosis and Prognosis
-
- Author / Creator
- Paul, Animesh Kumar
-
While it is very difficult to diagnose/prognosis psychiatric disorders reliably, especially in early course, such early diagnosis/prognosis is critical for producing an effective treatment. This necessity has motivated many researchers to apply machine learning approaches to high-dimensional neuro-imaging data, to produce models that can produce accurate diagnosis, and prognosis. The machine learning problems are more challenging due to their small sample sizes and large feature sets. These challenges motivated us to explore various novel ways of applying machine learning methods for predicting the diagnosis and prognosis of psychiatric disorders.
We considered the following 3 tasks: (1) We built a classifier that can distinguish healthy subjects vs. Obsessive-compulsive disorder (OCD) patients. In the learning pipeline, we incorporated prior neurobiological knowledge in terms of using pre-defined brain atlases for feature extraction. The best model (ensemble logistic regression) achieved 80.3\% accuracy. We also demonstrated a way to transfer information across psychiatric diagnoses, e.g., schizophrenia (SCZ) to OCD. (2) We next explored ways to apply machine learned schizophrenia diagnostic model to identify first degree relatives (FDR) subjects with high schizotypy scores. Our empirical results found that FDRs of SCZ patients who were classified as schizophrenia by a diagnosis model, which was learned using SCZ patients and healthy subjects, had significantly higher ‘schizotypal personality scores’ than those who were not classified as schizophrenia. (3) We addressed the challenges of building a prognostic model for SCZ patients. Here, we dealt with two types of SCZ patients based on their treatment: antipsychotic medication vs. transcranial direct current stimulation (tDCS) treatment. Our success is limited -- achieved 63.77% accuracy from the deep transfer learning model -- in predicting treatment response for SCZ patients with antipsychotic treatment. On the other side, our proposed prior neurological knowledge method for SCZ patients with tDCS-treatment was able to provide 77.5% accuracy for predicting the treatment response.
-
- Graduation date
- Fall 2020
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.