A new framework for mental illnesses diagnosis using wearable devices aided by improved convolutional neural network.

Stress inherent in the modern world is considered one of the main causes of Mental Health Disorders (MHDs) that spread in every country around the world. These mental and behavioral problems primarily affect the mind and brain that change emotions and perception, especially if they are not diagnosed or treated early. MHD diseases are difficult to be distinguished from each other because they come in many forms with different severity of symptoms and different periods of suffering. A person's bioactivity can be measured by the wearable technology such as smart watches that become more advanced and widely spread. A new framework based on analyzing the motor activity data measured by smart watches is presented to diagnose mental illnesses such as schizophrenia and depression as well as analyze complex behavioral patterns. The framework encodes the behavioral time series data into image patterns using modified Markov Transition Field. These images have been processed using a modified Convolutional Neural Network and an attention pooling approach for the classification of depression and schizophrenia patients. The proposed system achieved an accuracy of 96.6% in schizophrenia and 94.85% in depression as a precision of 92.1% in schizophrenia and 96.77% in depression. The proposed system is able to analyze the motor activity of individual and able to diagnose the mental illness. The results of the proposed system are superior to those used in detecting mental health disorders.