The early diagnosis of Alzheimer's disease (AD) plays an essential role in patient care since it allows to take prevention measures, specially at the early stages, before irreversible brain damages are shaped. Although many studies have applied deep learning methods for computer-aided-diagnosis of AD obtaining significant results and high prediction accuracies, the neuroimages utilized in these researches were preprocessed and, in result, far different from those produced originally by the medical equipment. In this study, a structural magnetic resonance imaging classifier based on a convolutional neural network has been designed for the diagnosis of AD from non preprocessed neuroimages. Keras pre-trained models Inception-v3 and VGG-16 have been implemented using transfer learning techniques to classify AD from cognitive normal (CN) and mild cognitive impairment (MCI), its prodromal stage. The proposed approach is validated on the standardized structural MRI datasets from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) project. This approach achieves an accuracy of 62% and 36% on Inception-v3 and 55% and 38% on VGG-16 for AD vs CN and AD vs MCI vs CN classification respectively. The further improvement and research of this algorithms using pre-selected 2,5 or 3D images have the potential to provide a rapid, accessible and computational efficient data-driven assessment of Alzheimer Disease.
Python 3.7 procedures for each model written in jupyter notebooks.
Requirements in requirements.txt
Used conda environment and conda install package.
Add additional notes about how to deploy this on a live system
- Daniel Coll - Initial work - AlzheimerDiagnose