Deep Learning-Based Detection and Severity Classification of Alzheimer's Disease Stages Using a Custom CNN Model

It is a neurodegenerative disorder wherein the cognitive function is drastically affected, particularly in aging populations. An early and definitive diagnosis ensures an appropriate treatment strategy that improves patient outcomes. The present study is directed toward presenting a Custom Convolutional Neural Network model, Custom CNN, to classify the stages of AD with high precision using MRI data. The dataset is made up of 6,400 MRI images divided among four classes: Non_Demented, Very_Mild_Demented, Mild_Demented, and Moderate_Demented. The preprocessing done on the data was resizing, normalization, and data augmentation to make the model robust. In the Custom CNN, several convolutional layers were included, dropout for regularization, and SoftMax for multi-class classification. Evaluation Metrics: Some of the evaluation metrics that have been used in assessing the performance of the model are: Accuracy, Precision, Recall, F1-score, confusion matrix. It reaches the general accuracy of 98% in being quite reliable in the various classifications, with some difficulties only in discriminating conditions that are very similar. These findings point toward great potential for deep learning models in supporting early AD detection. Future work will focus on improving feature extraction and model transparency through explainable AI, therefore contributing to the development of more accessible and effective diagnostic tools across different healthcare settings.