Analysis of Gait Patterns to Facilitate Alzheimer's Disease Level Classification using Deep Learning

Abstract

Recently, with rapid growth of object detection technology based on deep learning, the technology has been applied in diverse industries. In this paper, we describe the case where the object detection is used in medical field. In detail, we analyzed behavior pattern of normal control (NC), mild cognitive impairment (MCI), Alzheimer’s disease (AD) subjects. To conduct a reliable behavior pattern experiment, we adjusted the hyper parameters of the existing deep running object models and applied the data augmentation, fine tuning, and transfer learning method to enhance intersection of union (IoU) of the models. Through the experiment, we selected our own object detection models and then conducted behavior pattern analysis of AD, MCI, and NC using selected our object detection model. For the behavior pattern analysis, we suggested center point movement and distance between center point and walking-line experiment. The movement experiment showed average movement of NC (0.245), MCI (0.259), and AD (0.281). In the second experiment, we got the distance degree of NC (0.169), MCI (0.228), and AD (0.423). The results of two experiments also showed that there is meaningful difference among NC, MCI, and AD through T-test and ANOVA (analysis of variance).