Background: Toe-walking (TW) is present in about 20-30% of individuals with autism spectrum disorder (ASD). In some individuals, TW can completely disappear over time without any intervention; in others, it can persist and cause physical problems. TW quantification tools can provide insights into TW trajectories and aid in planning possible interventions. Recently, a video-based coding protocol based on standardized video recordings was proposed to quantify TW in an ecological setting: an operator reviews the videos and computes the parameters. The gold-standard to analyse gait is optoelectronic or wearable motion capture. This is difficult to use in individuals with ASD and cognitive disability because of their non-acceptance. An automated markerless gait analysis system (MGAS) approach could overcome these limitations.

Objectives: This study aims: 1) to implement an MGAS approach to quantify TW and 2) to validate the MGAS approach by comparing it with the reference video-recording approach.

Methods: We conducted a preliminary study on 5 individuals with ASD diagnosed according to DSM-5 criteria. The diagnosis was confirmed through the Childhood Autism Rating Scale (CARS-2), and the intellectual disability was assessed through the Vineland-2. TW was quantified using a modified version of a validated TW structured video-based coding protocol based on standardized video recordings. The test was conducted without shoes, albeit with socks. It consists of transporting 1 object (e.g. puzzle piece, Lego®) from the therapist to the playing table situated 3 meters away and back again 10 times. The assessment was repeated in three days for each individual. Thus, 15 tests were available for analysis. Observing the video, a therapist calculated the percentage of toe steps with respect to the total number of performed steps. OpenPose is the automated MGAS used to analyze the video-recorded tests. A previous study demonstrated its validity. The intraclass correlation coefficient (ICC) was used to assess the reliability between the therapist's visual assessment and MGAS approaches in quantifying percent of toe steps. Also, the Bland-Altman method checks for systematic distortion of measurement values. The 95% Limits of Agreement (LOA) was calculated as 1.96 times the standard deviation (SD) of the difference between the two methods. A narrow LOA range confirms that the methods could be used interchangeably.

Results: The 5 individuals with ASD had a mean age of 9.62 years (SD 2.46) (age range: 7.5-13.2 years; male: 4/5). The mean CARS-2 total score was 37.3 (SD 7.84). The mean Vineland-2 total quotient score was 27.2 (SD 11.35). Two individuals with ASD out of 5 presented TW. The normal data distribution was confirmed (Shapiro-Wilk test p>0.05). The ICC value of the two approaches was 0.975 (excellent). The result of Bland-Altman analysis was shown in Figure 1: an upper LOA below 18.28 and a lower LOA above 29.11 were obtained.

Conclusions: The MGAS OpenPose used for the quantitative assessment of TW in individuals with ASD and intellectual disability showed excellent reliability compared to the reference therapist's visual assessment. Bland-Altman analysis suggests the possibility of clinical application. Further research is required to confirm the results of this preliminary study.

Notes: 

(1)Autism Research Unit, Villa Santa Maria Institute, Tavernerio, Como, Italy, (2)School of Medicine and Surgery, University of Milano-Bicocca, Milano, Milano, Italy, (3)PhD Program in Neuroscience, University of Milano-Bicocca, Milano, Milano, Italy, (4)Scientific Institute IRCCS “E.Medea”, Bosisio Parini, Lecco, Italy, (5)Autism Research Unit, Villa Santa Maria SCS, Tavernerio, Como, Italy, (6)Department of Design, Politecnico di Milano, Milan, Italy