Predicting medical specialty from text based on a domain-specific pre-trained BERT

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Natural Language Processing (NLP)
• Clinical Decision Support Systems

Background:

• The COVID-19 pandemic highlighted the need for efficient outpatient management and timely treatment initiation.
• Accurate diagnosis and referral are crucial for effective healthcare delivery.
• Artificial intelligence (AI) and NLP offer potential solutions for advanced clinical systems.

Purpose of the Study:

• To develop and evaluate an AI model for predicting the appropriate medical specialty based on patient-submitted medical questions.
• To enhance the accuracy and efficiency of patient routing within healthcare systems.

Main Methods:

• A medical specialty prediction model was developed using pre-trained bidirectional encoder representations from transformers (BERT).
• The model was fine-tuned on a dataset of patient medical questions and corresponding specialty labels (27 categories).
• Performance was evaluated against four deep learning NLP models using cross-validation and test set evaluation on a real-world dataset.

Main Results:

• The proposed BERT-based model demonstrated superior performance in predicting medical specialties compared to other NLP models.
• Case studies confirmed the model's utility and potential for visualization applications.

Conclusions:

• The developed AI model can significantly improve hospital patient management by providing accurate specialty recommendations.
• This facilitates more efficient patient flow and ensures patients are directed to the correct medical specialists.