Robust broad learning system with parametrized variational mode decomposition for schizophrenia diagnosis

Schizophrenia (SZ) is a significant mental disorder characterized by various neurophysiological and cognitive impairments. Early diagnosis remains challenging due to its reliance on symptom detection. However, advance signal processing algorithm is combined with machine learning technique for early detection of schizophrenia using electroencephalogram (EEG) signals efficaciously. To optimize results from biomedical signals, effective feature extraction (FE) and feature engineering are essential. In this study, parametrized variational mode decomposition (PVMD) is applied to electroencephalogram (EEG) signals to extract band-limited intrinsic mode functions (BLIMFs), which are selected using fuzzy dispersion entropy (FDE). The extracted BLIMFs are fed into deep stack autoencoder (DSAE) with a minimum reconstruction error, utilizing root mean square (RMS) as the cost function. We also demonstrate how to apply the robust broad learning system (RBLS) to classify neuro-disorders, comparing it with various broad learning system (BLS) methods for schizophrenia classification. Building on RBLS's success, we propose a novel VMD-based BLS (VMD-BLS) technique. To address VMD-BLS's limitations, we introduce a PVMD-DSAE based RBLS (PVMD-DSAE-RBLS). The effectiveness of PVMD-DSAE-RBLS is tested on three datasets, with results showing accuracies of 99.98%, 96.91% and 99.29% for the Poland, Kaggle, and Moscow datasets, respectively. The performance of the proposed PVMD-DSAE-RBLS method significantly outperforms compared to similar learning algorithms and state-of-the-art techniques. Finally, a reconfigurable high-speed field-programmable gate array (FPGA) embedded processor is implemented to design a computer-aided diagnosis (CAD) system, providing efficient automated diagnosis for schizophrenia patients.