Traditional autism screening often relies on structured questionnaires or clinical observation, which can sometimes delay early identification. A paper by Sumaiya Afroz Mila, Jeba Maliha, Md Rafiul Kabir, Ankan Ghosh, and Sandip Ray introduces a different approach, utilizing large language models (LLMs) to make early screening more intuitive and accessible.
“ScreeningPaL: LLM-NLP Enabled Early Autism Detection from Caregiver’s Free-Text Input” explores how LLMs and natural language processing can be used to support early autism screening by analyzing unstructured caregiver-written descriptions of their child’s behaviors and developmental patterns. By leveraging caregiver free-text narratives, the model can detect linguistic patterns associated with autism-related behavioral indicators. The framework uses transformer-based language models to process and classify caregiver responses, aiming to provide a scalable and accessible screening support tool that could help identify children who may benefit from earlier clinical evaluation.
The paper received the 3rd Place Student Paper Award in the American Medical Informatics Association (AMIA) Amplify Informatics Summit in May 2026 after a presentation by Mila, a PhD student advised by Dr. Sandip Ray. It was one of the only all-engineering research papers to receive an award at the summit, with most winners being from medical schools.
“The broader motivation behind this work is to improve accessibility to early screening resources, particularly for families who may face barriers to timely specialist assessment,” Mila said. “By using natural language descriptions that caregivers can provide in their own words, the system seeks to make screening more intuitive and potentially more widely deployable.”
