Table 2. Demographic information on children in the collected home videos.
We collected N = 193 (119 ASD, 74 non-ASD) home videos for analysis. We excluded 31 videos because of inadequate labeling or video quality. We used a randomly chosen 25 autism and 25 non-autism videos to empirically define an optimal number of raters. Video feature tagging for machine learning was then done on 162 home videos.
| Videos | N | NASD | Nnon-ASD | Mean age (SD) | ≤2 years | >2 years and ≤4 years | >4 years and ≤6 years | >6 years | Percent male, ASD | Percent male, non-ASD |
|---|---|---|---|---|---|---|---|---|---|---|
| Total | 193 | 119 | 74 | 4 years 4 months (2 years 1 month) |
24.4% (n = 47) | 33.7% (n = 65) | 25.9% (n = 50) | 15.5% (n = 31) | 39.38% (n = 76) | 23.32% (n = 45) |
| Excluded | 31 | 3 | 28 | 3 years 8 months (1 year 11 months) |
32.3% (n = 10) | 29.0% (n = 9) | 25.8% (n = 8) | 9.6% (n = 4) | 3.23% (n = 1) | 58.06% (n = 18) |
| Total videos used for analysis of all 8 classifiers | 162 | 116 | 46 | 4 years 4 months (2 years 2 months) |
22.8% (n = 37) | 34.5% (n = 56) | 25.9% (n = 42) | 16.7% (n = 27) | 67.2% (n = 78) | 56.5% (n = 26) |
| Subset of videos used to find minimally viable number of raters | 50 | 25 | 25 | 4 years 6 months (2 years 4 months) |
28% (n = 14) | 34% (n = 17) | 18% (n = 9) | 20% (n = 10) | 48% (n = 12) |
44% (n = 11) |
Abbreviation: ASD, autism spectrum disorder.