Optimizing Structural MRI for Alzheimer's Diagnosis: A Preprocessing Approach Using FSL

Alzheimer's disease (AD) is a common neurodegenerative disorder. Early detection of AD is crucial for early intervention and is difficult due to mild anatomical changes in the brain. Magnetic Resonance Imaging (MRI) can be of primary importance in the non-invasive diagnosis of these anatomical changes, particularly if used in conjunction with Artificial Intelligence (AI) models. However, raw MRI data often contain noise, irregularities, and structural inconsistencies that can negatively affect the accuracy of the AI model. In this research, we propose a preprocessing workflow that will enhance MRI quality before it is used to train AI models for AD prediction. The workflow includes reorientation, MRI registration using FLIRT, skull stripping using BET using the FMRIB Software Library (FSL) within the Windows Subsystem for Linux (WSL) framework. Subsequently, the images were sliced using Slicer software to generate 2D axial, coronal, and sagittal views. The preprocessed images are then compiled into HDF5 (.h5) format, suitable for machine learning applications, using an 80:20 split for training and testing. The proposed workflow ensures standardized, noise-free, and analysis-ready MRI data, which will help support reliable and interpretable AI models for the early diagnosis of AD.