Modeling schizophrenia disease based on the neuronal structure in the involved areas
چکیده
Schizophrenia is a serious mental disorder that affects all aspects of a person's life. In this study, we introduced a new method based on a fuzzy inference system (ANFIS) to classify the EEG signal of 14 patients with schizophrenia and 14 healthy individuals. First, based on previous studies, we selected 16 channels of EEG signals from the 19 available channels that had the most changes between the two groups. Each channel was windowed using 4 feature extraction methods of SpEn, ShEn, SpEn, and ALVSAC. These windows were subdivided into 5 frequencies regions: delta (2-4), theta (4.5-7.5), alpha (8.5-12.5), beta (13-30), and gamma (30-45) by the Butterworth filter. On each of these frequency bands, 4 feature extraction methods were implemented and the average of all windows in each of the signal channel areas was reported separately as the final feature. 46 features were selected from the 640 features with the highest accuracy, followed by classifiers of ANFIS, SVM and ANN and k-fold cross-validation with k = 5, the accuracy of their classifications for these features was 99.89%, 98.89% and 95.59%, respectively. Also, the frontal area and F4 channel with the sub frequencies regions of delta-SpEn and theta-SpEn with the most involvement were selected as the appropriate channel. By ANFIS classifier and k-fold cross-validation, frontal area and F4 channel were selected as the best channels. Next, to model person with schizophrenia, we attempted to collect a complete conceptual model that included 11 areas involved in the patient and 6 neurotransmitters connecting between areas.Next, to model a person with schizophrenia, we tried a complete conceptual model that collected most of the areas involved in the disease and the neurotransmitter between the areas. RBF neural networks with a number of 50 neurons were designed to quantitatively model single neurons in neuronal regions in the involved blocks. The F4 channel signal was used as the output of a healthy and patient mathematical model to learn the coefficients between the affected areas in the Glu_PFC region. After training in healthy model coefficients, the effect of neuronal death and active and inactive changes in neurotransmitter physiology in schizophrenia on neurons and healthy model coefficients were tuned to patient model coefficients based on patient F4 channel output. MSE results related to the comparison of healthy and patient model output and their actual signal were 0.55% and 0.37%, respectively. In the last stage, the effect of 9 common drugs on the target areas was implemented by changing the coefficients of the patient model, then the output signal of the treatment model was labeled healthy using ANFIS classification. Also, we proposed new drugs with increased acetylcholine activity in PFC and VTA regions, increased GABA activity in PFC, VTA, and DRN regions, and implemented on the treatment model for validation and received healthy labeling.