A new EEG-based schizophrenia diagnosis method through a fuzzy DL model

Schizophrenia disrupts behavioral and cognitive manifestations such as thinking, perception, and speech. Early diagnosis is crucial in treating and limiting the effects of the disease. This research introduces an innovative approach to automatically recognize schizophrenia by combining deep convolution models and fuzzy logic. The model structure is developed through 6 convolution layers and 1 softmax layer. For addressing uncertainty, the activation functions in the convolutional neural network (CNN) design are adopted using type-2 fuzzy concepts. The data is processed through generative adversarial networks to boost diagnostic accuracy further and reduce overfitting. This method is assessed for electroencephalogram signals captured from patients with schizophrenia and healthy controls, demonstrating an impressive accuracy rate of 99.05% in distinguishing between schizophrenia and healthy individuals. Moreover, the recommended approach undergoes external validation on an unseen electroencephalogram database, achieving a diagnostic accuracy of 96.42%, with 100% sensitivity and 92.85% specificity. These outcomes significantly outperform existing approaches and establish a new standard for integrating ML technologies in medical applications. The model also proves resilient against a broad range of noise conditions.