Developing a Chatbot to Support Individuals With Neurodevelopmental Disorders: Tutorial

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Neurodevelopmental Disabilities Research

Background:

• Families of individuals with neurodevelopmental disabilities (NDDs) face significant challenges accessing reliable online health information.
• Existing support, like in-person coaching, is limited, and general internet searches often prove insufficient for developing health literacy.
• Chatbots offer potential for accessible information delivery, but their application in specialized healthcare, such as NDDs, is underdeveloped.

Purpose of the Study:

• To develop and evaluate CAMI (Coaching Assistant for Medical/Health Information), a novel chatbot designed to provide families with trusted resources for neurodevelopmental disabilities (NDDs).
• To improve access to core knowledge and services relevant to individuals with NDDs and their families through an AI-powered conversational agent.
• To explore the integration of knowledge graphs and user-centered design in creating specialized health information chatbots.

Main Methods:

• Development of the CAMI chatbot using the Django framework, incorporating a knowledge graph to map entities and relationships within the NDD domain.
• Utilizing a combination of the Unified Medical Language System and expert-identified entities for NDD-specific named entity recognition from user queries.
• Enhancing entity detection through the enrichment of vocabulary with synonyms and lay language terms, informed by end-user collaboration.

Main Results:

• CAMI successfully provided relevant resources for the majority of user queries related to neurodevelopmental disabilities (NDDs).
• Enriching the chatbot's vocabulary with synonyms and lay terms significantly improved its ability to detect relevant entities in user input.
• The knowledge graph effectively connected symptoms, diagnoses, and resources, enabling the chatbot to offer comprehensive information.

Conclusions:

• CAMI represents the first chatbot specifically designed to offer resources for neurodevelopmental disabilities (NDDs), demonstrating the potential of conversational AI in this area.
• Engaging end-users and integrating their input with standard ontologies is crucial for enhancing the language recognition capabilities of health information chatbots.
• Knowledge graphs provide a powerful method for integrating complex medical information, with implications for developing specialized chatbots across various health domains and optimizing expert input.