Abstract
Background/Objectives: Autism spectrum disorders (ASD) is one of the most common neurodevelopmental disorders with an estimated incidence of 1/54. Endophenotyping in these patients is very complicated.
Hence 3D analysis ("gestalt" analysis) of facial traits is a novel promising method. Methods: In the presented work we performed a cluster analysis of sets of 3D scans of ASD patients (116) and controls (157) using euclidean and geodesic distances.
Results: In a clustering scenario, where all possible distances between selected landmarks were used, the control group is significantly, but not completely separated from ASD cases. Additionally, in the same scenario, the ASD group was partitioned into several sub-groups and open-mouth subjects tend to cluster together.
However, the resulting clusters exhibit significant differences in BMI implying that BMI is the main factor determining clustering structure. When using data cleaned from outliers with only a subset of all distances not correlating with BMI, the cluster structure observed above completely disappeared and the controls were not separated from the ASD cases.
This conclusion applies to both euclidean and geodesic distances. Conclusion: In this presentation, we want to emphasize the confounding effect of BMI on the results of cluster analysis and the necessity to carefully control this factor in similar studies.