Diagnosis of Schizophrenia using Kolmogorov Complexity and Sample Entropy

Electroencephalogram (EEG) is known to be the least expensive means to record cerebral activities and anomalies. One on one sessions between the patient and the expert and the expert's analysis on the patient's mental condition based on Positive And Negative Syndrome Scale (PANSS) is presently the diagnosis procedure performed to assess the presence of mental disorder and its severity. Schizophrenia (SZ) is a chronic psychological disorder that influences the patients feelings, behaviour and perception, normally resulting in delusions, auditory hallucinations. This paper discusses how EEG can be an useful tool in diagnosing SZ in patients by designing a computer-aided tool. Dataset involved in this study consists of EEG recordings from nine electrodes : Fz, FCz, Cz, FC3, FC4, C3, C4, CP3, CP4. Schizophrenia affects the cortical regions of the brain. This paper investigates that the readings from EEG electrodes Cz, FCz, C3, and FC3 connected closer to the cortical region can produce better results in diagnosing SZ when compared to the results obtained from applying the same algorithm on readings of all nine EEG electrodes. Features like Kolmogorov Complexity and Sample Entropy are extracted and are used by Resilient Back-propagation Neural network for training and to perform classification. Algorithm when applied on all nine electrodes resulted in 88.75% accuracy, sensitivity of 88%, specificity of 93% and precision of 81.2% whereas considering only four electrodes gives an accuracy of 91.25%, sensitivity of 90.8%, specificity of 93.2%and precision of 96%. The reduction in number of electrodes involved in analyses reduces the time required for computation, storage capacity as well as improvises the result. Hence, EEG readings from Cz, FCz, C3 and FC3 can be used in accurate diagnosis of SZ.