Comparing three algorithms of automated facial expression analysis in autistic children: different sensitivities but consistent proportions.
Liora Manelis-Baram, Tal Barami, Michal Ilan and 5 others
PMID 41068936WHAT IT FOUND
Automated software found no difference in how often autistic and non-autistic children made facial expressions during 45-minute assessments.
However, the three algorithms disagreed significantly on detection rates, meaning current tools are not yet reliable for clinical measurement.
Key findings
01There were no significant differences in the quantity of facial expressions (happiness, anger, disgust, fear, sadness, surprise) between autistic and control children, regardless of which of the three algorithms was used.
02The three algorithms showed poor agreement with each other; iMotions detected faces in significantly fewer frames than FaceReader and Py-Feat, and Py-Feat identified emotions in three to four times as many frames as the other two.
03The quantity of facial expressions was not significantly correlated with autism severity as measured by ADOS-2 scores.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
No comparison was made against manually annotated ground truth data, so the accuracy of the algorithms could not be validated, only their agreement. The study only included verbal children who could complete ADOS-2 modules 2 and 3, so results do not apply to non-verbal or minimally verbal children. The analysis only counted the presence of expressions, ignoring their quality, timing, or social appropriateness. The sample size, while large for this type of study, was still too small to capture the full heterogeneity of facial expression patterns in autism.
Declared interests
Funded by the Israel Science Foundation and the Azrieli Foundation. No commercial conflicts of interest were declared by the authors.
The easy way to misread this
Do not interpret the lack of difference in expression frequency as evidence that autistic children do not have non-verbal communication difficulties. The study only measured the quantity of expressions, not their quality or appropriateness, and the authors note that existing algorithms may not accurately detect the specific types of expressions used by autistic individuals.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →