A Lightweight Face Image–Based Auxiliary Detection Model for Autism Spectrum Disorder

Chenkai LiaoWenqiu Zhu^*

• Hunan University of Technology, Zhuzhou, China

Early diagnosis of Autism Spectrum Disorder (ASD) plays a crucial role in improving patients' quality of life. In recent years, face image–based ASD detection has attracted increasing attention as an auxiliary diagnostic approach. However, existing lightweight models still show limitations in capturing fine-grained facial features. To address this problem, this paper proposes a lightweight face image–based auxiliary detection model for ASD, termed MN-ASD. First, MobileNetV4-S is selected as the baseline framework. To enhance the model's ability to capture subtle facial details, the FReLU activation function is introduced, which strengthens spatial feature modeling. Second, to further improve the performance and accuracy of the model, the Coordinate Attention (CA) module is incorporated at the final stage. By jointly modeling positional and channel dependencies, the CA mechanism enables the model to better locate key facial regions and enhance global feature extraction. Finally, experimental results demonstrate that the proposed method achieves an accuracy of 93.33% on the Kaggle dataset, with precision, recall, and F1-score all outperforming other mainstream lightweight networks. These results validate the effectiveness of MN-ASD in capturing typical facial features of ASD and indicate its suitability for early auxiliary screening in resource-constrained scenarios.