Predicting Sarcopenia of Female Elderly from Physical Activity Performance Measurement Using Machine Learning Classifiers

Figures & data

Figure 1 Information of the IMU: (A) 3-axis of IMU and (B) the position of the IMU.

Abbreviations: IMU, Inertial measurement unit; VT, Vertical; ML, Mediolateral; AP, Anterior-posterior; L2, Second lumbar vertebra.

Figure 2 Experimental set-up for collecting physical data of the elderly.

Abbreviation: DXA, dual energy X-ray absorptiometry.

Table 1 Sarcopenia Stage Grouping, According to AWGS Criterion for Female

Stage	The Number of Subject	Mean Muscle Mass	Mean Muscle Strength	Mean Gait Speed
	–	< 5.4kg/m^2	< 18kg	< 1.0m/s
Normal	50	6.5 ± 1.8kg/m^2	26.2 ± 6.3kg	1.4 ± 0.2m/s
Sarcopenia	20	4.9 ± 0.3kg/m^2	17.2 ± 0.7kg	1.3 ± 0.1m/s
Severe sarcopenia	0	None

Note: Values are mean ± SD.

Figure 3 The framework of identifying the optimal classification model for predicting sarcopenia-stage.

Abbreviations: ACC_VT, vertical linear acceleration; ACC_AP, anterior-posterior acceleration; ACC_ML, mediolateral acceleration; ACC_RES, resultant acceleration; YAW, Vertical axis angular velocity; ROLL, anterior-posterior axis angular velocity; PITCH, mediolateral axis angular velocity; kNN, k-nearest neighborhood algorithm; SVM, support vector machines; NB, Naïve Bayes algorithm.

Table 2 Features of TUG

	Features	Definition
Time features	Total time	Total time on TUG
	Sit-to-stand time	Time on sit-to-stand for gait
	Gait time	Average time on forward gait and backward gait in TUG
	Mid-turn time	Time on rotation at return point
	End-turn time	Time on rotation to sit on a chair
	Stand-to-sit time	Time on stand-to-sit on a chair
Descriptive statistic features	Root mean square(RMS)	Arithmetic mean of the squares of a set of values
	Min	The smallest values
	Max	The greatest values

Figure 4 The example of identification the TUG sub-tasks using angular velocity.

Abbreviations: PITCH, mediolateral axis angular velocity; YAW, Vertical axis angular velocity.

Table 3 Features of 6mWT

	Features	Definition
GT	The number of step	Number of steps for 6 minute
	Step/s	Step per second
	Stride length	Distance between steps
	Gait distance	Walking distance for 6 minutes
	Average gait speed	Average walking speed for 6 minutes
GS	Step regularity	Symmetry between steps as identified by 3-axis acceleration signals for walking
	Stride regularity	Symmetry between strides as identified by 3-axis acceleration signals for walking
	Symmetry index	Gait symmetry index
HR	HR	Smoothness & rhythm of acceleration signals measured for walking
ApEn	ApEn	Regularity of acceleration signals measured for walking

Abbreviations: GT, gait parameter; GS, gait symmetry parameter; HR, Harmonic ratio; ApEn, Approximate entropy parameter.

Table 4 Feature Selection Datasets

	FALL	FTUG	F6mWT
Datasets	10 features	10 features	5 features
	20 features	20 features	10 features
	30 features	30 features	15 features
	40 features	40 features	21 features(All)
	50 features	50 features
	132 features(All)	111 features(All)

Abbreviations: F_ALL, dataset consisted of both TUG features and 6mWT features; F_TUG, dataset consisted of only TUG features; F_6mWT, dataset consisted of only 6mWT features.

Figure 5 Classification models performance depending on feature selection datasets: (A) Results of F_ALL, (B) Results of F_TUG, (C) Results of F_6mWT, and (D) Final results.

Abbreviations: F_ALL, dataset consisted of both TUG features and 6mWT features; F_TUG, dataset consisted of only TUG features; F_6mWT, dataset consisted of only 6mWT features; kNN, k-nearest neighborhood algorithm; SVM, support vector machines; NB, Naïve Bayes algorithm; F_ALL40features(kNN), the kNN classification model using 40 features on F_ALL; F_TUG40features(SVM), the SVM classification model using 50 features on F_TUG;; F_6mWT15features(kNN), the kNN classification model using 15 features on F_6mWT.