Predicting Neurological & Psychiatric disease from resting state EEG Analysis<\/p>\n
Quantitative EEG analysis and predictions using deep learning tools have attracted a lot of attention due to the powerful repercussion that they can have on diagnosis of neurological and psychiatric disease.<\/p>\n
Herein I present the analysis of a publicly available, closed eyes, resting state EEG study of 36 individuals diagnosed with Alzheimer’s disease (AD group), 23 diagnosed with Frontotemporal Dementia (FTD group) and 29 healthy subjects (CN group). A Nihon Kohden EEG 2100 clinical device was used, with 19 scalp electrodes according to the 10-20 international system and 2 reference electrodes (A1 and A2) placed on the mastoids for impedance check. Data are publicly available at OpenNeuro: https:\/\/openneuro.org\/datasets\/ds004504\u00a0<\/a><\/p>\n I used standard band pass filtering, common average referencing and ICA to remove any artefacts, before segmenting the data in 5 sec epochs with 1 second overlaps and removing any bad epochs based on noise levels (A). <\/p>\n I then use this dataset to perform a three-label classification analysis via deep learning (DL) techniques (B) by building a variant of a hybrid model architecture of sequences of 1D convolutional units (Convolutional Neural Networks [CNNs], Inception modules) followed by series of layers of gated recurrent units (Recurrent Neural Networks [RNNs], GRU modules). Such structures have been utilized before and have been shown to be successful in discerning differences in pathological EEG (see Roy, Kornek & Harrer 2018, ChronoNet: A Deep Recurrent Neural Network for Abnormal EEG Identification arXiv, https:\/\/lnkd.in\/eQ9pt4Ts<\/a> ) <\/p>\n I show that after only 50 rounds of training with gradient descend learning protocols using an Adam optimizer with labelled epoched data (and a development set) a mean 88.6 % accuracy of classification prediction is achieved when the model is presented with unknown data that it has never seen before (test set C, D). The data show that the model most often confused the FTD condition for AD (18% of epoch case) but scored highly in classifying correctly CN (91%) or AD (94%) individuals. Had that been a two-class problem, the results would have been staggering, but this is not the point here!<\/p>\n Generalization of deploying EEG DL models in clinical practice can be easily augmented via behavioral or blood data to increase the multimodality of the training and achieve high level of confidence in diagnosis.<\/p>\n The take home message is this: We live in 2025: Let us bring a fresh objective to neurology and psychiatry when it comes to disease diagnosis, staging and future prognosis. Patients and their families depend on us. <\/p>\n The model is written in python and has been deposited in my GitHub. <\/p>\n","protected":false},"excerpt":{"rendered":" Predicting Neurological & Psychiatric disease from resting state EEG Analysis Quantitative EEG analysis and predictions using deep learning tools have attracted a lot of attention due to the powerful repercussion that they can have on diagnosis of neurological and psychiatric disease. Herein I present the analysis of a publicly available, closed eyes, resting state EEG […]<\/p>\n","protected":false},"author":1,"featured_media":1994,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_eb_attr":"","footnotes":""},"categories":[1],"tags":[],"class_list":["post-1993","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","entry","has-media"],"_links":{"self":[{"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/posts\/1993","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1993"}],"version-history":[{"count":11,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/posts\/1993\/revisions"}],"predecessor-version":[{"id":2005,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/posts\/1993\/revisions\/2005"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=\/wp\/v2\/media\/1994"}],"wp:attachment":[{"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1993"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1993"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/antoniosdougalis.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1993"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}