Predictive analysis for Alzheimer's diagnosis through data mining techniques

Alzheimer's disease represents one of the most significant challenges for contemporary healthcare, as there is still no effective cure for it. Magnetic Resonance Imaging (MRI) techniques have contributed to the diagnosis and prediction of its progression, but they require time and specialized skills for image analysis. Therefore, the use of deep learning techniques is crucial in analyzing large amounts of MRI images with high accuracy for early detection and prediction of the disease progression. In the following work, we focused on feature extraction from multiple sources and their integration to improve the accuracy of Alzheimer's diagnosis. Three distinctive methodologies have been developed. The first one utilizes a Feed Forward Neural Network (FFNN) with features extracted from models like ResNet50 and DenseNet201 with and without the application of Principal Component Analysis (PCA). The second methodology combines features from both models, with and without the application of PCA. Finally, the third methodology combines features from the models with those extracted from manual techniques like Discrete Wavelet Transform (DWT), Local Binary Pattern (LBP), and Gray Level Co-occurrence Matrix (GLCM) by applying PCA before or after feature combination.