Abstract:
Alzheimer's disease (AD) is an irremediable, irrecoverable brain illness without proper treatment. Therefore, AD recognition is essential for precluding and dominating its advancement. The conventional method considers the patient history, MRI, and empirical tests to detect AD. This method is effective, but it lacks accuracy. The primary objective is to create an efficient framework for timely AD identification from MRI images. This work combines a Convolutional Neural Network (CNN) with the K-nearest neighbor (KNN) to detect AD from 6400 MRI images. The dataset consists of four types of AD MRI images such as Moderate Demented (MOD), Mild Demented (MID), Very Mild Demented (VMD), and Non-Demented (ND). The dataset was split into training (80 percent) and validation (20 percent) sets. The CNN was utilized to extract the informative features automatically from the MRI dataset images, and these extracted features are used to train and validate the KNN model. The CNN-KNN framework is accessed utilising the Receiver Operating Characteristic (ROC) curve, stratified K-fold, Cohen's Matthews Correlation Coefficient (MCC) and Kappa Coefficient (CKC). This method achieved average 99.58 percent accuracy, 99.63 percent precision, 99.31 percent recall, 99.43 percent F1-score, 99.31 percent MCC, and 0.9931 CKC. Furthermore, three deep CNNs (DCNNs), ResNet50, VGG16, and MobileNetV2, were employed to detect AD and compare their performance with CNN-KNN. The proposed CNN-KNN method's performance has significantly improved compared to the DCNN and literature.
