Design and Development of an Automated Electroencephalogram Signal Processing and Diagnostic Support Architecture...

Electroencephalogram (EEG) signals play a vital role in analyzing neurological activity and diagnosing various neurological disorders. With the rapid growth of the Internet of medical things (IoMT), EEG-based diagnostic systems have become more interconnected and scalable. However, early and accurate diagnosis of neurological diseases remains challenging due to the intrinsic complexity, high dimensionality, and noise-prone nature of EEG signals. This study aims to design an automated, robust, and intelligent diagnostic support framework for reliable early detection of neurological diseases.

This paper proposes the design and development of an automated electroencephalogram signal processing and diagnostic support architecture employing advanced deep learning models (EEG-DSS-SMCNN). EEG signals were collected from Alzheimer's disease and Frontotemporal Dementia datasets. Initially, a Fast Desensitized Kalman Filter Algorithm (FDKFA) is applied for preprocessing to remove noise, artifacts, and signal distortions while ensuring data normalization. Subsequently, a Graph Fractional Fourier Transform (GFFTT) is employed to extract discriminative features from time, spatial, and channel domains. The extracted features are then fed into a Self-Modulating Convolutional Neural Network (SMCNN) to perform automatic disease detection and classification. Furthermore, the network parameters were optimized using an elephant clan optimization (ECO) algorithm to enhance classification accuracy and model robustness.

Experimental evaluation demonstrates that the proposed EEG-DSS-SMCNN framework significantly outperforms existing state-of-the-art methods. The proposed approach achieves accuracy improvements of 6.42%, 6.58%, and 8.27% compared to RT-EEG-DL, EEML-DLT-EEG, and EDDL-ED-EEG methods, respectively, confirming its superior diagnostic performance.

The results highlight the effectiveness of integrating advanced signal preprocessing, graph-based feature extraction, and deep learning within a unified diagnostic framework. The proposed EEG-DSS-SMCNN model efficiently handles noisy EEG signals and captures complex neurological patterns, making it suitable for early-stage disease diagnosis in IoMT-enabled healthcare environments. The optimized SMCNN further enhances classification reliability, suggesting strong potential for real-time clinical decision support and scalable neurological disease monitoring systems.