Application of Convolutional Neural Network Based on ResNet18 for Alzheimer Disease Classification

Abstract

Alzheimer's disease is a form of progressive dementia that significantly impacts the quality of life of patients and their families. Early detection based on Magnetic Resonance Imaging (MRI) can support faster and more accurate diagnosis, but manual classification requires high expertise and is subjective. This study aims to develop an Alzheimer's MRI image classification model using a Convolutional Neural Network (CNN) based on ResNet18 with transfer learning to classify data into four categories: Mild Demented, Moderate Demented, Non-Demented, and Very Mild Demented. The MRI dataset was processed through pre-processing involving 128×128 grayscale conversion, pixel intensity normalization, and class balancing using class weighting. The model was trained using the Adam optimizer (lr=0.0001) with Early Stopping (patience=7) over 50 epochs. Evaluation using the validation set showed that the model achieved high accuracy for the Non-Demented class. The result indicates that ResNet18 with transfer learning can achieve an accuracy of 94.4%, making this model an effective approach for medium-scale classification of Alzheimer's MRI images.