Early Diagnosis of Alzheimer's Disease Using Informative Features of Clinical Data

Diagnosing Alzheimer's disease (AD) is usually difficult, especially when the disease is in its early stage. However, treatment is most likely to be effective at this stage; bringing an advantage in improving the life of patients, diagnosis process. After years of research, still little is known about its detailed mechanism. The AD patients undergo different physical examinations, brain scans, and laboratory tests etc. that require them to physically visit the medical center multiple times. Such visits further result in each patient's massive data stored for clinical diagnosis. This elevates the possibility of using informative rich variables from this data for the early detection of AD with the help of Machine Learning (ML) techniques. However, the previously proposed models endure a number of limitations which place strong barriers towards the direct applicability of such models for accurate prediction. A number of classifiers have been utilized in the literature but none of the previous work utilized the two major categories of variables namely clinical diagnosis and clinical judgment. In this paper, we utilize these two categories of data and perform a comparative evaluation of the predominant machine learning algorithms in terms of prediction accuracy, precision, recall (AUC) and training time. Our experimental results revealed that Bayesian based classifiers improve AD detection accuracy and allows the meaningful interpretation of predictive model which assists in early prognosis of AD for each patient.