Enhanced Alzheimer’s Disease Prediction through Advanced Imaging: A Study of Machine Learning and Deep Learning Approaches

Abstract

The early detection and effective treatment of Alzheimer’s disease increasingly benefit from advanced imaging techniques. This study evaluates machine learning (ML) and deep learning (DL) methods to find Alzheimer’s disease using a dataset of 12,000 images, divided into four groups of 3,000 images each, and processed with Histogram of Oriented Gradients (HOG) feature representations. In the ML analysis, we applied Principal Component Analysis (PCA), Standard Scaling, and Stratified K-Fold cross-validation across four algorithms. The Support Vector Machine (SVM) achieved an accuracy of 79%, Random Forest reached 67% Decision Tree obtained 51%, and Logistic Regression recorded 65%. For the DL approach, we utilized data augmentation and trained a Convolutional Neural Network (CNN), achieving 87% accuracy. Transfer learning with the VGG16 model resulted in the highest accuracy of 91%. Our comparative analysis demonstrates that the VGG16 model significantly outperforms traditional ML algorithms and a custom CNN model in finding Alzheimer’s disease from medical images, highlighting the considerable potential of advanced DL techniques, particularly transfer learning models, in enhancing diagnostic accuracy for Alzheimer’s disease.

Topics & Concepts