Hemavathi U., Durai S
One of the common neurodegenerative disorders is called Alzheimer’s disease (AD), which gradually reduces human memory, thinking ability, and cognition. Deep learning (DL) approaches play a vital role in early AD prediction because they do not require manual feature learning like traditional machine learning (ML) techniques. However, further improvement is still possible in terms of detection accuracy by reducing training time and minimizing overfitting. This research study develops a DL technique for early AD prediction and includes a pre-trained convolutional neural network (CNN) based feature extraction. The proposed technique consists of three phases: (1) preprocessing, (2) segmentation, and (3) classification. Initially, the system collects the input MRI images from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. Then, contrast-limited adaptive histogram equalization (CLAHE) is employed to enhance the contrast of the gathered data from the dataset. The model then employs the Mahalanobis distance-based k-means (MDKM) algorithm to segment the skull portions from the input images. Finally, the system utilizes the osprey-optimized densely connected network 169 (O2DenseNet169) to classify AD, with hyperparameters optimally selected using the osprey optimization algorithm (OOA). The proposed technique achieves 99.44% accuracy, demonstrating the effectiveness of O2DenseNet169 in automatically learning biomarkers from images to classify AD.