Medical Specialty Recommendations by an Artificial Intelligence Chatbot on a Smartphone: Development and Deployment

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• The COVID-19 pandemic has disrupted patient-provider contact, complicating access to specialist care.
• Smartphone-compatible AI chatbots can classify symptoms and recommend medical specialties, offering a potential solution.

Purpose of the Study:

• To develop a deep learning-based natural language processing (NLP) pipeline for contactless medical specialty recommendation.
• To create a smartphone-compatible AI chatbot for symptom classification and specialist referral.

Main Methods:

• Collected and cleansed 51,134 sentences of symptom-label data for training deep learning models.
• Trained and validated various models including LSTM and BERT, evaluating performance using precision, recall, F1-score, and AUC.
• Developed a web-based AI chatbot using the Alpha framework for user-friendly symptom-based specialty recommendations.

Main Results:

• The Bidirectional Encoder Representations from Transformers (BERT) model achieved the highest AUC (0.964) and F1-score (0.768).
• The Long Short-Term Memory (LSTM) model with embedding vectors was selected for the chatbot due to resource efficiency and performance (AUC 0.965, F1-score 0.739).
• An Alpha version of the AI chatbot was successfully deployed for desktop and smartphone use.

Conclusions:

• AI chatbots leveraging deep learning NLP models are valuable for telemedicine, enabling rapid, contactless medical specialty recommendations.
• The developed chatbot supports patients and primary care providers by facilitating efficient access to appropriate specialists based on symptoms.