Ohmoto Yoshimasa, Terada Kazunori, Shimizu Hitomi, Kawahara Hiroko, Iwanaga Ryoichiro, Kumazaki Hirokazu
Department of Behavior Informatics, Faculty of Informatics, Shizuoka University, Shizuoka, Japan.
Department of Electrical, Electronic, and Computer Engineering, Faculty of Engineering, Gifu University, Gifu, Japan.
Front Psychiatry. 2024 Oct 17;15:1464285. doi: 10.3389/fpsyt.2024.1464285. eCollection 2024.
Research supporting the presence of diverse motor impairments, including impaired balance coordination, in children with autism spectrum disorder (ASD) is increasing. The one-legged standing test (OLST) is a popular test of balance. Since machine learning is a powerful technique for learning predictive models from movement data, it can objectively evaluate the processes involved in OLST. This study assesses machine learning's effectiveness in evaluating movement in OLST for predicting high autistic trait.
In this study, 64 boys and 62 girls participated. The participants were instructed to stand on one leg on a pressure sensor while facing the experimenter. The data collected in the experiment were time-series data pertaining to pressure distribution on the sole of the foot and full-body images. A model to identify the participants belonging to High autistic trait group and Low autistic trait group was developed using a support vector machine (SVM) algorithm with 16 explanatory variables. Further, classification models were built for the conventional, proposed, and combined explanatory variable categories. The probabilities of High autistic trait group were calculated using the SVM model.
For proposed and combined variables, the accuracy, sensitivity, and specificity scores were 1.000. The variables shoulder, hip, and trunk are important since they explain the balance status of children with high autistic trait. Further, the total Social Responsiveness Scale score positively correlated with the probability of High autistic trait group in each category of explanatory variables.
Results indicate the effectiveness of evaluating movement in OLST by using movies and machine learning for predicting high autistic trait. In addition, they emphasize the significance of specifically focusing on shoulder and waist movements, which facilitate the efficient predicting high autistic trait. Finally, studies incorporating a broader range of balance cues are necessary to comprehensively determine the effectiveness of utilizing balance ability in predicting high autistic trait.
支持自闭症谱系障碍（ASD）儿童存在多种运动障碍（包括平衡协调受损）的研究越来越多。单腿站立测试（OLST）是一种常用的平衡测试。由于机器学习是一种从运动数据中学习预测模型的强大技术，它可以客观地评估OLST中涉及的过程。本研究评估机器学习在评估OLST运动以预测高自闭症特征方面的有效性。
在本研究中，64名男孩和62名女孩参与。参与者被要求面对实验者单腿站在压力传感器上。实验中收集的数据是与脚底压力分布和全身图像相关的时间序列数据。使用具有16个解释变量的支持向量机（SVM）算法开发了一个识别高自闭症特征组和低自闭症特征组参与者的模型。此外，还针对传统、提议和组合的解释变量类别构建了分类模型。使用SVM模型计算高自闭症特征组的概率。
对于提议变量和组合变量，准确率、敏感度和特异度得分均为1.000。肩部、髋部和躯干变量很重要，因为它们解释了高自闭症特征儿童的平衡状态。此外，社会反应量表总分与各解释变量类别中高自闭症特征组的概率呈正相关。
结果表明，通过使用视频和机器学习评估OLST中的运动来预测高自闭症特征是有效的。此外，它们强调了特别关注肩部和腰部运动的重要性，这有助于高效预测高自闭症特征。最后，需要纳入更广泛平衡线索的研究，以全面确定利用平衡能力预测高自闭症特征的有效性。
