Deep EEG-Based AlexNet Convolutional Neural Network for Automated Dementia Classification

The purpose of this study was to compare the electroencephalograms (EEGs) of five patients with vascular dementia (VD), fifteen patients with moderate cognitive impairment (MCI) due to a stroke, and fifteen healthy controls (CN) as they performed a working memory (WM) task. As a hybrid pre-processing strategy, we used independent component analysis (ICA) and the wavelet transform (WT) to clean up the EEG data. It is proposed to use a deep learning approach to analyze noninvasive scalp EEG recordings to automatically detect and classify dementia patients. The method suggested here involves calculating the spectrogram of the EEG time series and then visualizing the resulting spectral profiles as 3 dimension colour images (3-D spectrogram images). To extract valuable features from these images and carry out the associated multi-classification tasks. A Convolutional Neural Network (CNN) built on the framework of AlexNet was first employed. next, the output features of the modified AlexNet model's convolutional and pooling layers are classified using the k-nearest neighbour (kNN) method. The research showed that when the deep learning (DL) pre-trained AlexNet CNN model was combined with transfer learning, and kNN machine learning (ML) classifier had the best output with an accuracy of 80.70%, whereas the pre-trained AlexNet DL model's results were 70.18 %. According to the results, an AlexNet and kNN classifier combination produced the best results, and the suggested approach can prognosticate the severity of dementia more accurately than baseline methods with a nearly 10.6% increase in accuracy. In addition to potentially assisting in the detection of patients with VD and MCI, these results provide credence to the idea that deep learning systems could be useful in improving the discrimination of CN individuals, patients with MCI, and those with vascular dementia.