Visual feature analysis on selective appetite in individuals with autism spectrum disorders

Abstract

Background: Individuals with autism spectrum disorders (ASD) experience more severe selective eating problems than their neurotypical peers. Identifying the causes of selective eating behavior poses a considerable challenge, even for caregivers. Accurate identification of the underlying causes of this behavior is essential for developing interventions aimed at overcoming dysfunctional, unbalanced diets. However, studies that meticulously identify the causes of selective eating behaviors are scarce. This investigation aims to explore the differences in preferences for sunny-side-up eggs between individuals with ASD and those with typical development (TD), focusing on the factors influencing their likes and dislikes through a systematic analysis of visual features.

Method: Thirty-nine individuals with ASD (mean age, 23.4 ± 4.7 years; 82% men) and fifty individuals with TD (mean age, 22.2 ± 1.3 years; 64% men) participated in this study. We used a total of 50 images of sunny-side-up eggs as visual stimuli. Using Non-negative Matrix Factorization and Decision Tree analysis, factors associated with visual preferences for sunny-side-up eggs were identified.

Data and results: We could identify factors associated with visual preferences for sunny-side-up eggs. Subsequent linear regression analysis provided insight into how these visual features delineate preference boundaries between liked and disliked foods, with noteworthy distinctions emerging between the ASD and TD groups.

Conclusions: This study provides novel insights into the visual determinants of food preferences in individuals with ASD through systematic analysis of image features. Our findings indicated the potential to predict preferences while elucidating the causes of selective eating behaviors, thereby offering solutions for individuals with ASD.

Fig 1. Photography setup for sunny-side-up egg images.

The setup consists of a light box with LED illumination and a calibrated camera system connected to a laptop computer. A white fabric draped over the entire setup blocks external light to ensure controlled lighting conditions. A color checker chart is included in each image for color calibration. All images were captured under these standardized conditions to ensure consistent illumination across samples.

Fig 2. Classification of sunny-side-up egg images based on the questionnaire.

The results showed a difference in the distribution of visual preferences between individuals with ASD and TD. In the top row, there are 8 images that both the ASD and TD groups liked, 12 images that only the TD group liked, 19 images that both the ASD and TD groups disliked, 1 image that only the ASD group disliked, and 10 images that did not belong to either of the two groups. All images are ordered according to the mean preference scores of the ASD group. ASD, autism spectrum disorder; TD, typical development.

Fig 3. Visualization of base vectors extracted by NMF.

NMF, non-negative matrix factorization.

Fig 4. Scatter plot of sunny-side-up egg images based on NMF Factor 1 and Factor 6.

The horizontal axis represents Factor 1 and the vertical axis represents Factor 6. Blue markers represent ratings by the ASD group, and orange markers represent ratings by the TD group. Circles indicate images rated as “Liking,” whereas triangles indicate images rated as “Disliking.” Solid lines show the linear regression boundaries between “Liking” and “Disliking” ratings for each group (blue for ASD and orange for TD). Dotted lines represent the 95% confidence intervals for these boundaries. ASD, autism spectrum disorder; TD, typical development; NMF, non-negative matrix factorization.

Fig 5. Scatter plot of sunny-side-up egg images based on NMF Factor 1 and Factor 8.

The horizontal axis represents Factor 1 and the vertical axis represents Factor 8. Blue markers represent ratings by the ASD group, and orange markers represent ratings by the TD group. Circles represent images rated as “Liking,” whereas triangles represent images rated as “Disliking.” Solid lines show the linear regression boundaries between “Liking” and “Disliking” ratings for each group (blue for ASD and orange for TD). Dotted lines represent the 95% confidence intervals for these boundaries. ASD, autism spectrum disorder; TD, typical development; NMF, non-negative matrix factorization.