Wrist Actimetry Biomarker Development of Paretic Upper Limb Use in Post Stroke Patients for Ecological Monitoring

14 Background 15 In post-stroke patients it is unclear which wrist actimetry biomarkers to use to estimate the 16 degree of upper limb hemiparesis. The objective of this study was to develop a general and 17 objective framework for monitoring hemiparetic patients in their home environment via 18 different biomarkers based on 7 days of actimetry data. A secondary objective was to use all of 19

these biomarkers to better understand the mechanism for potential non-use of the paretic upper 20 limb. 21

22
Accelerometers were worn continuously for a period of 7 days on both wrists of 10 post-stroke 23 hemiparetic patients as well as 6 healthy subjects. Various wrist actimetry biomarkers were 24 calculated, including the Jerk ratio 50 (JR50, cumulative probability that the Jerk Ratio is 25 between 0 and 0.5), absolute and relative amounts of functional use of movements of the upper 26 limbs (FuncUse and FuncUseR) and absolute and relative velocities of the upper limbs during 27 functional use (VUL and VULR). For each biomarker, the values of stroke and healthy groups 28 were compared. The correlations between all the biomarkers were studied. 29

30
We studied 10 participants with mild-to-moderate chronic hemiparesis and 6 healthy control   the measurement of acceleration in the three spatial directions. The accelerometers were 114 recovered at the end of the 7 days to extract the data using the OmGui software provided by 115 Axivity. The data were sliced day by day to obtain daily acceleration data values. The data were 116 then saved in csv format so they can be read by any programming language. 117

118
Data processing was done using the python 3.7 programming language. The numpy and scipy 119 libraries are notably used for numerical calculation operations (derivation, frequency analysis). 120 The scipy library allows the application of a low pass filter with a cut-off frequency of 10Hz in 121 order to remove noise. The magnitude of the acceleration vector (SVM: scalar vector 122 magnitude) is then calculated for each time step of the two actimeters (via the acceleration data 123 at a given time t : ax(t); ay(t); az(t)). 124 Pan et al., [Pan,2020] showed that the jerk ratio (JR) is sensitive to the degree of upper limb 137 mobility. The jerk ratio is defined as the ratio of the jerk amplitude of the paretic (non-dominant) 138 limb to the sum of the jerk amplitude of the paretic (non-dominant) limb and the nonparetic 139 The probability density function is normalised to give a total probability distribution of 1. 147 Following the work of Pan et al, [Pan et al, 2020], the jerk ratio 50 (JR50) was calculated. This 148 metric corresponds to the cumulative probability that the JR is between 0 and 0.5. A JR50 value 149 greater than 0.5 suggests a preponderant non-paretic (dominant) arm mobility. 150

2) Forearm Elevation angle and speed 151
In quasi-static condition, the calculation of the angle of elevation of the forearm with respect to 152 the gravity vector takes the form of equation 6, following the trigonometric laws: 153 It is then possible to obtain the angular velocity of elevation by the time 155

186
In this study, 6 healthy (3 women) and 10 post-stroke patients (6 women) participated. The 187 characteristics of the patients and healthy subjects are summarised in Table 1. 188     The aim of the study was to calculate multiple wrist actimetry biomarkers of stroke patients 299 over a 7-days period in their home environment and then determine optimal biomarkers to 300 monitor functional paretic arm use (FuncUse). We performed, to our knowledge, the first study 301 in stroke patients that calculated over an extended 7-days period multiple functional movement 302 biomarkers via two simple and lightweight wrists worn accelerometers, and compared these 303 values with values acquired in a healthy population. Accordingly, we derived new actimetry 304 biomarkers, in particular, we were able to calculate average elevation speed of execution of  This suggests that the stroke patient studied here maintain a relatively normal amount of non-337 paretic UL movement average. 338 The Jerk Ratio appears to reflect a ratio of the amount of movement in a given time frame 339 between the two limbs. While this ratio is balanced in healthy subjects, it shows a slight 340 imbalance in stroke subjects. These results show that there is a significantly higher probability 341 that stroke patients perform less movement, both functional and non-functional, with their 342 paretic limb than with their non-paretic limb when compared with the healthy population. 343 Furthermore, the study of correlations between the different biomarkers seems to show a 344 decreasing exponential relationship between the FuncUseR and the JR50. This suggests that 345 depending on the degree of deficit of the stroke patients, the two biomarkers would be 346 complementary in establishing a diagnosis. Indeed, the FuncUseR seems to be more sensitive 347 for patients with upper limb behavior similar to healthy subjects, whereas the JR50 seems to 348 be more sensitive for subjects with significant hemiparesis (Figure 4.D). Furthermore, the 349 results showed that stroke patients had significantly lower average execution speeds of 350 functional movements than healthy subjects. It should be noted that the measured elevation 351 speeds seem to correspond to the values of the literature [Lacquaniti et al, 1982]. It is 352 interesting to note that there is a very strong positive correlation between the FuncUseRatio 353 and the VULR in healthy subjects but not in strokes patients. Finally, the principal component 354 analysis showed that the PC1 allows to differentiate with sufficient sensitivity the actimetric 355 results of healthy and hemiparetic subjects. We also see that the moderate hemiparetic 356 subjects have the lowest PC1 values. 357 In order to define a functional movement of the upper limbs we have arbitrarily chosen to define 358 an amplitude of elevation of the arm of more or less 30°. However, a large proportion of stroke 359 patients show uncontrolled flexion of the healthy elbow when walking. This phenomenon is 360 called "associated reaction" and may have an influence on the results of our study [Kahn et al, 361 2020]. This choice remains arbitrary and it would be necessary to explore the evolution of the 362 FuncUseR as well as the functional movement quantities as a function of this elevation 363 amplitude parameter. In particular, we would expect to observe no significant difference 364 between post-stroke and volunteers' subjects for functional movements of plus or minus 10° of 365 elevation. Instead, the difference would tend to increase with the amplitude of the movement. 366 It would then be possible to identify an angular amplitude threshold value for each patient and 367 thus to obtain a new parameter allowing to better identify the patient deficiency.

Conclusions 393
This study comparing healthy and post-stroke subjects found significant differences in 394 calculated actimetric biomarkers between healthy and post-stroke subjects. While the healthy 395 subjects had an upper extremity functional use ratio close to 1, the post-stroke subjects had