32240
Automatic Analysis of Temporal Patterns of Head Movement Discriminate between ASD Outcome Groups
Recent work has shown the importance of quantifying head movement differences in children with and without ASD to better characterize the ASD phenotype (Martin et. al, 2017). However, these analyses relied on aggregated data and did not examine the temporal dynamics of head movement in children with and without ASD. Further, little is known about the head movement patterns in siblings of children with ASD. In this study, we performed a temporal pattern analysis to determine whether the sequential patterns of head movement differed by family history of ASD and ASD outcome.
Objectives:
A computational framework was implemented to extract the temporal patterns of head movement tracking data. We then examined whether these temporal patterns differentiated between ASD outcome groups.
Methods:
Fifty-four participants were 2.5-6.5-year-old children (mean=4.72 years, SD=1.14 years) with (high-risk, n=33) and without (low-risk, n=21) an older sibling with ASD. ASD diagnoses were confirmed for low- and high-risk children by a licensed clinical psychologist. We examined differences in head movement patterns in three independent groups: children with ASD (n=21), children at high-risk for ASD (n=12, HR/NoASD), and low-risk children (n=21, LR/NoASD). Children were video-recorded while watching a 16-minute video containing social and nonsocial stimuli. Three dimensions of rigid head movement—pitch, yaw, and roll (Figure 1) —were tracked using an automatic person-independent tracker (Zface). The temporal patterns were extracted from the head movement tracking data of multiple motion feature dimensions. The categorical timeline allocation process identified the imbalance of the temporal pattern distributions between groups using mean discriminative ratios (mDR). A higher mDR indexes greater differential occurrence of head movement patterns (their differential presence in one group but not another).
Results:
The categorical timeline allocation results show differential occurrence of sequential patterns between groups. Sequential patterns based on pitch, yaw, and roll differentiated ASD and HR/NoASD children, with mDRs of .91, .90, and .94, respectively. Sequential patterns based on pitch, yaw, and roll differentiated LR/NoASD and HR/NoASD children, with mDRs of .90, .92, .94, respectively. Sequential patterns based on pitch, yaw, and roll moderately differentiated ASD and LR/NoASD children, with mDRs of .52, .76, .64, respectively (Table 1). There were 7 patterns of pitch and yaw and 8 patterns of roll that were observed in the ASD but not the HR/NoASD group. There were 7 patterns of pitch and yaw and 8 patterns of roll that were observed in the LR/NoASD but not the HR/NoASD group. There were 2 patterns of pitch and roll and 5 patterns of yaw that were observed in the ASD but not the LR/NoASD group (Table 1).
Conclusions:
The presented analysis framework identified potential diagnostically-relevant head movement motion patterns in ASD. Temporal patterns extracted from head movement tracking data show the ability to discriminate between ASD outcomes. Many temporal patterns appear in one group but are absent in the other, suggesting that these unique patterns distinguish between groups. Together, the discriminative power of temporal patterns and their ability to distinguish between groups indicate the potential of these automatic methods for diagnostic utility in ASD.