Page 98 - 南京医科大学自然版

第44卷第12期
·1712 · 南 京 医 科 大 学 学 报 2024年12月

Experiment 1 Experiment 2 Experiment 3
1.0 1.0 1.0
ACC=0.763 ACC=0.723
n=20 n=135
0.8 0.8 0.8
ACC ACC=0.907 ACC ACC
0.6 0.6 0.6
n=40
0.4 0.4 0.4
0 50 100 150 200 0 40 60 80 0 50 100 150
Feature number Feature number Feature number

Experiment 4 Experiment 5 Experiment 6
1.0 1.0 1.0

0.8 0.8 0.8
ACC ACC ACC ACC=0.793
0.6 ACC=0.810 0.6 0.6 n=40
n=30 ACC=0.840
0.4 0.4 n=95 0.4
0 20 40 60 0 100 200 300 0 10 20 30 40
Feature number Feature number Feature number
Experiment 7 Experiment 8 Experiment 9
1.0 1.0 1.0

0.8 0.8 0.8
ACC ACC ACC
0.6 0.6 0.6
ACC=0.770 ACC=0.797
ACC=0.703
0.4 n=170 0.4 0.4 n=200
n=55
0 100 200 0 50 100 150 200 0 200 400 600 800
Feature number Feature number Feature number
Experiment 10 Experiment 11 Experiment 12
1.0 1.0 1.0

0.8 0.8 0.8
ACC ACC ACC
0.6 0.6 0.6
ACC=0.867 ACC=0.887
ACC=0.873
n=45 n=300
0.4 n=255 0.4 0.4
0 200 400 500 0 25 50 75 100 125 0 250 500 750 1 000 1 250
Feature number Feature number Feature number
Taking experiment 1 as an example，when only the features extracted from experiment 1 were used for modeling，the best feature set was deter⁃
mined by selecting features in ascending order based on their P⁃values. The performance of the observed model was then verified by modeling and cross⁃
checking. The experiment 1 had the highest the accuracy（ACC）of 0.907 when the number of features reaches 40，with a corresponding kappa value of
0.817.
图2 各UPDRSⅢ指定动作的运动参数模型
Figure 2 The motor parameter models for each designated movement in UPDRSⅢ

试 者 工 作 特 征（reciever operating characteristic， 动作的数据信息（表4）。其中，言语动作模型、下半
ROC）曲线如图3。 身动作中站立、行走、后拉试验等可触发全部传感
该多点可穿戴设备设置10个传感器位点，采集 器位点，上半身的动作只能触发 1~2 个传感器收集
包括言语、面部表情、手指拍打、手掌运动、前臂运 信号，结合表3及表4可发现，动作模型中触发的传
动、前臂回旋动作、脚趾拍地实验、两脚灵敏度测 感器位点越多，收集的运动参数量越多，其准确性
试、起立、5 m 折返走、后拉实验、双手平举、指鼻等 与一致性越高。