Visualization of deep networks trained for bipolardisorder classification by using fnirs measurements

Oğuzhan, Babacan
Deep learning applications have achieved impressive performances on many medi-cal problems such as classification of disorders, effects of a treatment or unspottedsymptoms of a disease, etc. While modern deep learning progress is impressive insuch areas, genuine understandings of its working principles are not clear. For thatmatter, the term black box has often been associated with deep learning algorithms.The majority of previous studies have concentrated on networks’ successes and havecomputed their performances in terms of accuracy levels. However, this thesis fo-cuses on disintegrating the internal working mechanisms of neural networks into in-tuitive and understandable components. It makes them easy to understand and tointerpret from medical experts’ perspectives. With this purpose in mind, pre-trainedConvolutional Neural Networks and Residual Neural Networks are utilized by usingtime-series neuroimaging data, i.e. Functional Near-Infrared Spectroscopy (fNIRS)measurements, belonging to two classes, namely healthy and bipolar, and their visu-alization outputs are attained. Since these outputs are complex time-series data, theyare analyzed by statistical methods such as chi-square and t-tests so that the intrinsicfeatures of healthy and bipolar subjects specific to their classes are obtained. Results are compared with previous medical studies and are analyzed so that potential reasonsbehind the classification results are provided. The contribution of this thesis is pro-viding an inference about visualization outcomes of different neural networks, whichare trained for the bipolar disorder classification using fNIRS data. Therefore, thisstudy tries to fill the void between medical researchers and deep learning experts.
Citation Formats
B. Oğuzhan, “Visualization of deep networks trained for bipolardisorder classification by using fnirs measurements,” M.S. - Master of Science, Middle East Technical University, 2021.