Categorization of Behaviors of Alzheimer’s Disease Patients Using Deep Learning

Abstract

As the population aging, the number of elderly people with dementia is also increasing. Republic of Korea is also expected to become a super-aged society with 14.3% of the population aged 65 or older in 2018 and 32.8% in 2040 and 41.0% in 2060. The number of people with dementia will also increase from about 760,000 in 2018 to 1.96 million in 2040, and 2.95 milion in 2060. At the same time, the social costs of dementia will increase from 11.7 trillion won in 2013 to 43.2 trillion won in 2050. The total annual cost of supporting each person (18.51 million won) increases as the severity of dementia increases. This paper ues RetinaNet to address the issue of early diagnosis of dementia and classification of severity. Following the pre-process of stitching the subject images through the affine transformation, the proposed model detects the subject within the image and recognizes and analyzes the walking pattern of the detected subject and classifies that they are normal or dementia patient and the severity of the dementia. Normal and dementia patients were recruited from March 2016 to October 2018 and the procedure for collecting the subject data for experiments was reviewed by the Gwangju Institute of Science and Technology's Bioethics Committee, and executed after screening and neurological test. For walking pattern analysis, we compared change of total walking time for each tasks and groups. Secondly we experimented that observe locomotion which mean each step took a similar time.