Early Detection of Alzheimer's Disease Using Artificial Intelligence Based on CSF Biomarkers & MRI Images

Alzheimer is a degenerative disease. It affects the memory of the person and his cognitive ability, which becomes worse over time, depriving him completely of leading a normal life while being entirely dependent on others. There is still no cure for Alzheimer's disease, however, its rate of decline can be slowed with proper medication and therapy, especially if it was detected at an early stage. Moreover, the clinical examination alone is incapable of providing definite early diagnosis due to the confusion with normal aging symptoms. Therefore, it's important to find a way to improve the early detection of the disease allowing a considerable delay in the disease's progression and preserving the patient's cognitive ability for longer with the adequate treatment. In this study, we propose 2 methods to enhance the early detection of Alzheimer's disease: the use of cerebrospinal fluid (CSF) biomarkers and MRI images. The CSF biomarkers dataset will be fed into multiple machine learning and deep learning models which will be trained and tested to classify the data into Alzheimer's disease (AD), Mild cognitive impairment (MCI) and control. In addition, two MRI datasets are used: sagittal and axial. The aim is to detect whether multi-MRI view model can provide better accuracy at diagnosing Alzheimer's disease than single view model. A deep learning CNN algorithm is used to classify the datasets into healthy controls, mild demented, and moderately demented. Thus, each dataset will be preprocessed and trained separately then combined and trained furthermore. The results obtained from classifying the CSF biomarkers samples are 99.22% using decision tree classifier, 62.72% with logistic regression, 99.38% with random forest, 99.44% for KNN, 74.5% for SVM and 60.33% Naïve Bayes along with 84.27% for CNN model. As for the MRI, the testing accuracy reaches 90% using longitudinal MRI images, 92% for axial images and 94% for both data combined.