Reconocimiento de patrones comunes en Parkinson y Alzheimer aplicando técnicas de Inteligencia Artificial

Abstract

CONDITIONING OF PROJECT PUBLICATION. Neurodegenerative conditions such as Alzheimer’s disease and Parkinson’s disease can be difficult to identify at an early stage, as they often share similar clinical signs and underlying pathophysiological mechanisms. The objective of this project is to identify common structural patterns in magnetic resonance imaging (MRI) scans of patients with AD and PD using artificial intelligence techniques. For this purpose, MRI data from public databases such as ADNI and PPMI were used and subjected to preprocessing, standardization, and data augmentation procedures. Three deep learning models were trained and evaluated: a CNN Dual-Branch model, a 3D Vision Transformer, and a hybrid CNN-Transformer model. The results showed that the hybrid CNN-Transformer achieved the best balance between accuracy, generalization capability, and interpretability, outperforming the other evaluated architectures. In addition, the use of Grad-CAM made it possible to identify the brain regions most relevant for classification, facilitating the interpretation of the results by healthcare professionals. Finally, a web application was developed that integrates the selected model, allowing users to upload, analyze, and interact with MRI scans in a visual and intuitive manner. Keywords: Artificial Intelligence, Magnetic Resonance Imaging, Deep Learning, Neurodegenerative Diseases