Unified Deep and Machine Learning Hybrid Models for Alzheimer’s and Mild Cognitive Impairment Detection

Abstrak

Alzheimer’s disease (AD), the most common form of dementia, is characterized by progressive neurodegeneration, leading to memory loss and cognitive decline. Recent studies have reported annual conversion rates from amnestic Mild Cognitive Impairment (MCI) to probable AD. With the advent of Magnetic Resonance Imaging (MRI)-based analysis, advancements in machine learning (ML), particularly deep convolutional neural networks (CNNs), have transformed the diagnostic landscape of AD. However, earlier approaches often struggled to accurately distinguish between different MCI stages. To address this limitation, a deep neural network (DNN) model was developed, employing an enhanced artificial neural network (ANN) architecture to classify individuals into three categories: mild Alzheimer’s dementia, MCI, and normal cognition. The proposed DNN model, trained on a Kaggle dataset, achieved an exceptional accuracy of 0.99. In comparison, conventional classifiers such as Naïve Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA) achieved accuracies of 0.97, 0.96, 0.96, and 0.96, respectively. Meanwhile, K-Nearest Neighbors (KNN) attained 0.83, Random Forest (RF) achieved 0.95, and Logistic Regression (LR) reached 0.93. Hybrid models combining DNN with SVM and DT (DNN-SVM and DNN-DT) yielded accuracies of 0.79 and 0.64, respectively. These findings highlight the importance of selecting models that balance interpretability with computational efficiency. Overall, this study provides valuable insights into the strengths and limitations of various classification techniques, enabling informed decisions for different datasets and clinical objectives in Alzheimer’s disease diagnosis.