ChatASD: LLM-Based AI Therapist for ASD

Xiaoyu Ren, Yuanchen Bai, Huiyu Duan, Lei Fan, Erkang Fei, Geer Wu, Pradeep Ray, Menghan Hu, Chenyuan Yan, Guangtao Zhai

Abstract: LLMs have performed significantly in the medical field. While they cover a broad range of topics including internal and surgical diseases, and mental health issues like depression, their depth in specific professional domains, especially Neurodevelopmental Disorders (NDDs) like Autism Spectrum Disorder (ASD), is limited and prone to errors. It is evident that user-friendly, cost-effective, patient, knowledgeable, rational, and interactive LLMs could be an excellent tool, i.e., play a role in autism awareness, diagnosis and treatment. However, the current understanding of autism, the lack of datasets and innovative methods limit this tool’s potential. Therefore, in this paper, we conduct the first large-scale study in medical LLMs for autism. The first bilingual autism knowledge dataset with approximately 4500 entries is constructed, including multi-dimensional information about autism (e.g., education, treatment, inclusivity, etc.), real-case diagnostics, and easily confused concepts. Moreover, a LLM for autistic families called ChatASD is introduced, supporting bilingual knowledge dissemination and auxiliary diagnosis. Additionally, a LLM-based diagnostic and treatment pipeline for autistic patients called ChatASD Therapist is proposed, supporting bilingual dialogue and facial video generation. Our dataset and LLM-based tools represent a novel attempt to interact directly with autism patients and their families, providing inspiration for the continued exploration of diagnostic tools for ASD and other NDDs. The constructed database will be available at: https://github.com/DuanHuiyu/ChatASD.