Deep learning-based artificial intelligence imaging probes for Alzheimers disease

Abstract

Brain medical imaging is a main diagnosis method for Alzheimers disease (AD). But the method relies on the physicians manual analysis which is subjective and time consuming. In recent years, artificial intelligence (AI) technology has been widely applied in clinical diagnosis. This thesis is about the deep learning model to be designed to realize the computer-aided diagnosis of medical images. A model of densely connected network (DenseNet) as an AI technology, automatically learns the semantic features related to AD diagnosis on the brain MRI images from ADNI data. At the same time, for solving the limited medical image samples problem, the effective transfer learning technology was applied in the experiment. The final model result achieves 90.8% accuracy, 82.2% sensitivity and 96.1% specificity on the diagnostic task of AD, and the diagnostic accuracy is better than prevailing methods. Besides 80.4% accuracy, 52.2% sensitivity, and 84.8% specificity are achieved in the task of distinguishing progressive from stable MCI patients. This method can provide more accurate diagnosis results of Alzheimers disease expected for the clinical early auxiliary diagnosis.