Das-Alz: Alzheimer's Disease Classification Using Downscaled MRI Scans and a Reinforced Convolutional Neural Network

Low-resolution (LR) and high-resolution (HR) images are critical in medical imaging, with HR images often preferred for their detailed information. However, HR MRI scans require significant processing power, posing challenges in computational efficiency. Super-resolution techniques have been developed to enhance LR images, but they often depend on high-quality data and advanced processing capabilities. One type of neurodegenerative illness is Alzheimer's disease (AD), where a precise and timely diagnosis is essential for successful intervention. Despite the importance of HR MRI scans in making the prediagnosis, their processing demands are high. To address these challenges, this paper proposes the DAS-Alz model, a reinforced convolutional neural network designed to classify AD cases using downscaled MRI scans. Even at a lower resolution, DAS-Alz demonstrates the ability of deep learning to extract important information from MRI data, highlighting the potential to overcome computational constraints. The model incorporates a multi-head attention mechanism layer to enable parallel processing, thereby improving efficiency. The DAS-Alz model achieved outstanding accuracy in both training (99.34%) and validation (99.84%), effectively classifying AD from downscaled images. These results suggest that DAS-Alz may play a significant role in AD prediagnosis, potentially leading to improved patient outcomes.