Detecting Autism Spectrum Disorder with Deep Eye Movement Features

TL;DR

Autism spectrum disorder (ASD) diagnosis can benefit from eye-movement biomarkers, which exhibit discrete, short-term fixation patterns poorly suited to global attention in Transformer models. We propose Discrete Short-Term Sequential (DSTS), a CNN-based framework that captures local eye-movement dynamics using a two-layer Conv1d embedding, followed by a classifier. To enhance discriminative learning and handle data imbalance, DSTS integrates Class-aware Representation (CaR) via a Multisample Similarity Loss and an Imbalance-aware (Ia) mechanism, combining losses as $\mathcal{L}_{total} = \lambda \mathcal{L}_{CaR} + (1 - \lambda) \mathcal{L}_{Ia}$ with $\lambda = 0.25$. Across eight eye-movement datasets, DSTS outperforms Transformer-based and traditional methods in ACC and F1 while using fewer resources, highlighting the value of modeling local, discrete gaze patterns for ASD detection and suggesting a shift away from long-range attention for such data.

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by deficits in social communication and behavioral patterns. Eye movement data offers a non-invasive diagnostic tool for ASD detection, as it is inherently discrete and exhibits short-term temporal dependencies, reflecting localized gaze focus between fixation points. These characteristics enable the data to provide deeper insights into subtle behavioral markers, distinguishing ASD-related patterns from typical development. Eye movement signals mainly contain short-term and localized dependencies. However, despite the widespread application of stacked attention layers in Transformer-based models for capturing long-range dependencies, our experimental results indicate that this approach yields only limited benefits when applied to eye movement data. This may be because discrete fixation points and short-term dependencies in gaze focus reduce the utility of global attention mechanisms, making them less efficient than architectures focusing on local temporal patterns. To efficiently capture subtle and complex eye movement patterns, distinguishing ASD from typically developing (TD) individuals, a discrete short-term sequential (DSTS) modeling framework is designed with Class-aware Representation and Imbalance-aware Mechanisms. Through extensive experiments on several eye movement datasets, DSTS outperforms both traditional machine learning techniques and more sophisticated deep learning models.

Figures (3)

• Figure 1: Gaze Pattern Comparison between TD and ASD Participants. The numbers denote the sequential order of fixations, and the circle size corresponds to the duration of each fixation.
• Figure 2: Workflow of DSTS. Eyemovement data is processed through convolutional layers, integrating class-aware and imbalance-aware mechanisms, before passing the features to the classifier for prediction.
• Figure 3: Comparison of time consumption and parameter count across methods on the Speaking and Walking1 datasets.