[4T1-OS-6c] OS-6

[4T1-OS-6c-03] Comparison of time series prediction of physical balance between ASD and typically developing children using LSTM networks

〇Yuna Takeda¹, Maresuke Seta¹, Yoshimasa Ohmoto¹, Hirokazu Kumazaki², Ryoichiro Iwanaga², Akira Imamura², Hitomi Shimizu², Kazunori Terada³ (1. Shizuoka University, 2. Nagasaki University, 3. Gifu University)

Keywords:ASD, Screening, LSTM

Diagnosing Autism Spectrum Disorder (ASD) is time-consuming and requires quantitative methods. However, it is difficult to obtain consistent data from ASD children because of their diverse symptoms. Moreover, ASD affects motor coordination and posture control. Therefore, we hypothesized that ASD and typically development (TD) children have different motor coordination abilities. We used LSTM to build a model that predicts the temporal changes of the body during coordinated movements, such as one-leg standing. We compared the model predictions with the actual data of children at high risk for ASD and at low risk for ASD and found significant differences in the ankle angle variability, which correlated with the SRS-2 scores.