Detecting outliers of pursuit eye movements: a preliminary analysis of autism spectrum disorder
A new outlier analysis method identifies 7 out of 18 ASD adults via unique eye movement patterns.
A research team from Japan, led by Emiko Shishido, has published a novel study on arXiv that shifts the paradigm for analyzing autism spectrum disorder (ASD) using eye-tracking. Instead of comparing group averages, their method performs an 'outlier analysis' on smooth pursuit eye movements (SPEM) during a slow Lissajous pursuit task. They recorded data from 18 adults with ASD and 39 typically developed (TD) individuals, deriving an 'outlier score' based on the Mahalanobis distance from a TD normative distribution. The score was calculated from a feature vector—optimized via Principal Component Analysis (PCA)—comprising temporal lag (Δt) and spatial deviation (Δs). An individual was statistically defined as an outlier if their score exceeded √10 (approximately 3.16σ).
While the TD group showed a low outlier rate of 5.1%, the ASD group demonstrated a significantly higher prevalence of 38.9% (7 out of 18 participants), with a binomial P-value of 0.0034. The mean outlier score was also significantly elevated in the ASD group (3.00 ± 2.62) compared to the TD group (1.52 ± 0.80). Crucially, this method captured extreme individual deviations even when conventional mean-based comparisons showed limited sensitivity. The authors conclude that this approach successfully visualizes the high degree of idiosyncratic atypicality in ASD oculomotor control, providing a sensitive metric for the disorder's inherent heterogeneity.
The study's preliminary findings, published on March 24, 2026, as arXiv:2603.22705, suggest this outlier-focused framework could serve as a baseline for identifying clinical subtypes of ASD. By moving beyond group averages to quantify individual pathological signatures, the research offers a new lens for understanding neurodevelopmental conditions. The team has made supplementary material available on GitHub, and the full paper is accessible via arXiv, marking a step toward more personalized diagnostic tools in neuroscience and psychiatry.
- The study found a 38.9% outlier rate in the ASD group (7/18) vs. 5.1% in the TD group, a statistically significant difference (P = 0.0034).
- The method uses a Mahalanobis distance-based 'outlier score' from PCA-optimized features of temporal lag (Δt) and spatial deviation (Δs), with a threshold of √10 (≈3.16σ).
- It captures individual atypicalities in smooth pursuit eye movements that conventional group-mean analyses miss, offering a potential tool for subtype identification.
Why It Matters
This research pioneers a personalized, data-driven approach to autism diagnosis, moving beyond group averages to detect individual neurological signatures.