Seleção automatizada de séries temporais inerciais do teste de sentar e levantar: um estudo metodológico

Abstract

This is a quantitative research, of an exploratory nature, conducted through a survey, subjects participated in the study, of both sexes and who were able to perform the 30-second sit and stand test. The objective of this study was to create an algorithm capable of automatically segmenting and identifying a sit-to-stand cycle and analyzing the similarity between the curves obtained by the smartphone and curves previously recorded by kinematography. Thus, the study was divided into two stages: stage (i) creation of the template through kinematic recordings of acceleration from the sit-to-stand test; step (ii) automated identification of the cycles recorded by the smartphone, and finally, analyze the similarity between the curves through two metrics of similarity, cross-correlation and Euclidean distance. A total of 3749 cycles were segmented, and only 3492 were considered for analysis. The preliminary result showed that there was no significant difference between the cycle counting methods (p=0.96) and that the mean similarityof the cycles studied with the template was euclidean distance (DE de 40.2 ± 8.29) e de cross-correlation (CC de 0.64 ± 0.13). The correlation between CC and DD metrics was inverse and -0.81 (p <0.0001). The cutoff points established from the cumulative distribution returned similarity indicator values related to the percentage above or below the cutoff. For example, for a cutoff of 80%, CC 0,71±0,06 (20% of curves above this value) and DE 35.3±3.99 (80% of curves below this value) were obtained. When the cutoff point was raised to 90%, CC 0,74±0,05 (10% of curves above this value) and DE 33.4±3.85 (10% of curves below this value) were obtained.