Prediction of Psychiatric Disorders Using Deep Learning

Traditional approaches that rely on neurophysiological signals have been ineffective in diagnosing psychiatric problems because people tend to hide their true emotions. To address this, our research suggests an enhanced EEG-based diagnosis model using Deep Learning (DL) and Machine learning approaches. We sought to improve the diagnosis of psychiatric disorders by implementing with DL and machine learning models on EEG data, which has encouraging implications for future medical developments. We focused on six primary and nine specific illnesses utilizing a sample of 945 people, comprising 850 patients and 95 healthy persons. Quantitative EEG data including power spectral density (PSD) and many frequency bands were examined while the subjects were at rest. Long short-term memory (LSTM), K nearest neighbors (KNN), Convolution neural networks (CNN) techniques are used in proposed research for prediction. A CNN-LSTM hybrid model binary classification was carried out. It is noteworthy that all the suggested models, including the KNN, used complete band features to achieve 96.83% accuracy for obsessive-compulsive disorder. Notably, the accuracy of the CNN-LSTM models in predicting obsessive-compulsive disorder was 97.88%. EEG has a great deal of promise as a diagnostic tool for mental illnesses since it can enhance patient outcomes and care in the psychiatric profession.