Biosignal comparison for autism assessment using machine learning models and virtual reality.

Clinical assessment procedures encounter challenges in terms of objectivity because they rely on subjective data. Computational psychiatry proposes overcoming this limitation by introducing biosignal-based assessments able to detect clinical biomarkers, while virtual reality (VR) can offer ecological settings for measurement. Autism spectrum disorder (ASD) is a neurodevelopmental disorder where many biosignals have been tested to improve assessment procedures. However, in ASD research there is a lack of studies systematically comparing biosignals for the automatic classification of ASD when recorded simultaneously in ecological settings, and comparisons among previous studies are challenging due to methodological inconsistencies. In this study, we examined a VR screening tool consisting of four virtual scenes, and we compared machine learning models based on implicit (motor skills and eye movements) and explicit (behavioral responses) biosignals. Machine learning models were developed for each biosignal within the virtual scenes and then combined into a final model per biosignal. A linear support vector classifier with recursive feature elimination was used and tested using nested cross-validation. The final model based on motor skills exhibited the highest robustness in identifying ASD, achieving an AUC of 0.89 (SD = 0.08). The best behavioral model showed an AUC of 0.80, while further research is needed for the eye-movement models due to limitations with the eye-tracking glasses. These findings highlight the potential of motor skills in enhancing objectivity and reliability in the early assessment of ASD compared to other biosignals.

[Virtual reality and digital biomarkers: a clinical tool for early autism diagnosis]. / Realidad virtual y biomarcadores digitales: una herramienta clínica para el diagnóstico del autismo.

INTRODUCTION: Autism Spectrum Disorder (ASD) is a neurodevelopmental condition which traditional assessment procedures encounter certain limitations. The current ASD research field is exploring and endorsing innovative methods to assess the disorder early on, based on the automatic detection of biomarkers. However, many of these procedures lack ecological validity in their measurements. In this context, virtual reality (VR) shows promise for objectively recording biosignals while users experience ecological situations. METHODS: This study outlines a novel and playful VR procedure for the early assessment of ASD, relying on multimodal biosignal recording. During a VR experience featuring 12 virtual scenes, eye gaze, motor skills, electrodermal activity and behavioural performance were measured in 39 children with ASD and 42 control peers. Machine learning models were developed to identify digital biomarkers and classify autism. RESULTS: Biosignals reported varied performance in detecting ASD, while the combined model resulting from the combination of specific-biosignal models demonstrated the ability to identify ASD with an accuracy of 83% (SD = 3%) and an AUC of 0.91 (SD = 0.04). DISCUSSION: This screening tool may support ASD diagnosis by reinforcing the outcomes of traditional assessment procedures.

Introducción: El Trastorno del Espectro Autista (TEA) es un trastorno del neurodesarrollo, y sus procedimientos tradicionales de evaluación encuentran ciertas limitaciones. El actual campo de investigación sobre TEA está explorando y respaldando métodos innovadores para evaluar el trastorno tempranamente, basándose en la detección automática de biomarcadores. Sin embargo, muchos de estos procedimientos carecen de validez ecológica en sus mediciones. En este contexto, la realidad virtual (RV) presenta un prometedor potencial para registrar objetivamente bioseñales mientras los usuarios experimentan situaciones ecológicas. Métodos: Este estudio describe un novedoso y lúdico procedimiento de RV para la evaluación temprana del TEA, basado en la grabación multimodal de bioseñales. Durante una experiencia de RV con 12 escenas virtuales, se midieron la mirada, las habilidades motoras, la actividad electrodermal y el rendimiento conductual en 39 niños con TEA y 42 compañeros de control. Se desarrollaron modelos de aprendizaje automático para identificar biomarcadores digitales y clasificar el autismo. Resultados: Las bioseñales reportaron un rendimiento variado en la detección del TEA, mientras que el modelo resultante de la combinación de los modelos de las bioseñales demostró la capacidad de identificar el TEA con una precisión del 83% (DE = 3%) y un AUC de 0.91 (DE = 0.04). Discusión: Esta herramienta de detección puede respaldar el diagnóstico del TEA al reforzar los resultados de los procedimientos tradicionales de evaluación.