Empirical Analysis for Earlier Diagnosis of Alzheimer’s Disease Using Deep Learning
MetadataShow full metadata
With the aging of contemporary society, cognitive impairment and dementia in the elderly, mainly Alzheimer's disease (AD), have become increasingly serious. As a multifactor, multistage, and clinical syndrome with concomitant diseases, senile cognitive impairment will take progress to irreversible dementia after clinical symptoms appear, eventually lead to death. Alzheimer's disease is currently irreversible, effective treatments lack in clinical practice. The development of a patient's status will go through several stages, so early diagnosis is essential. Early intervention of Alzheimer's disease can effectively slow down the disease progression while reduce the burden on patients' families and our society. This thesis introduces a method based on deep learning for early diagnosis and screening AD. The method is to slice a 3D magnetic resonance image of a human brain so as to generate a two-dimensional image, then we use an object detection network Faster R-CNN to detect the atrophy of the hippocampus region of human brain to realize the diagnosis of AD. A new network is modified and optimized based on VGG16 as the basic network of Faster R-CNN to extract feature maps and obtain 100% high-precision detection of AD samples. At the same time, 97.67% of the detected image accuracy is obtained for the validation set.