A Deep Learning Approach for Multi-Class Classification of Alzheimer’s Disease Using MRI Imaging

Abstract

Early diagnosis is essential for effective treatment and patient care, as Alzheimer's disease is a progressive neurological disorder that has a significant impact on memory, cognitive function and overall brain function. In this work, an improved deep learning approach is developed to predict Alzheimer's disease (AD) from MRI brain images in the early stages. This study uses publicly available data sets like ADNI and MRI data sets from Kaggle that contain images categorized as Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. To boost data quality and model generalization, a thorough preprocessing pipeline was developed, including the resizing, normalization, data augmentation, and noise reduction of images. For baseline comparison, traditional machine learning models like SVM, Random Forest and KNN were used with the extracted features using PCA. Deep learning models however, showed superior performance, especially Convolutional Neural Networks (CNN) as they could automatically extract complex features. In addition, transfer learning models like VGG16, ResNet50 and EfficientNet-B0 were used to gain benefits from the pre-trained knowledge, and the ResNet50 model was found to be the most efficient model. Another possibility, which was also explored, was to have a hybrid system that would use CNN along with traditional classifiers, but this did not yield significant improvements. Various optimizers and hyperparameter tuning techniques were used extensively. The proposed CNN model outperformed the other models with an accuracy of 98.7% and the transfer learning model with 98.0% accuracy.The proposed CNN model had the highest accuracy of 98.7% and the transfer learning had an accuracy of 98.0% which was good. The system was implemented by developing a web application in Flask that allowed the physicians to upload the MRI images and get the diagnosis results in real-time. The findings underscore the potential of deep learning methods in enhancing diagnostic accuracy at early clinical stages and aiding clinical decision-making in detecting Alzheimer's disease.