Smart Insole-Based Classification of Alzheimer’s Disease Using Few-Shot Learning Facilitated by Multi-Scale Metric Learning
Younghoon Jeon, Jaeyong Kang, Byeong C. Kim, Kun Ho Lee, Jong‐In Song, Jeonghwan Gwak
Abstract
Alzheimer’s disease is a progressive brain disorder, and mild cognitive impairment is a predisposing stage to it. Although various diagnostic methods have been proposed, they are difficult to use periodically due to pain and cost. In addition, small-sized medical dataset problems due to cost and ethical issues are challenging for artificial intelligence. Therefore, we propose multilevel gait experiment paradigms as diagnostic tools and a few-short learning-based diagnostic model facilitated by metric learning to solve the small data problems. In this study, two types of gait datasets were acquired from 69 subjects using a smart insole for several performance evaluation experiments. Experimental results showed that our proposed model improved with increasing difficulty of paradigm and outperformed conventional deep metric learning methods.