WhyMedQA: Enhanced biomedical why question answering using transfer learning approach

Area of Science:

• Natural Language Processing (NLP)
• Biomedical Informatics
• Artificial Intelligence

Background:

• Large language models (LLMs) struggle with specialized medical vocabulary and accuracy in biomedical question answering (QA).
• Existing NLP models lack the domain-specific understanding required for complex medical concepts.

Purpose of the Study:

• To introduce WhyMedQA, a novel biomedical QA system.
• To enhance LLM performance in the medical domain using a specialized transformer-based model.

Main Methods:

• Developed a transformer-based model on the BART architecture with domain-specific modifications.
• Fine-tuned the model on BioASQ8 and PubMedQA datasets.
• Incorporated additional layers to improve understanding of biomedical terminology and context.

Main Results:

• The proposed model achieved lower training and validation losses compared to baseline models.
• WhyMedQA demonstrated higher BLEU and ROUGE scores, indicating improved response quality.
• The model requires significantly fewer parameters, enhancing suitability for resource-constrained environments.

Conclusions:

• WhyMedQA offers a specialized and efficient solution for biomedical question answering.
• The model's advancements in medical NLP can potentially reduce decision-making errors and improve patient outcomes.
• This work highlights the effectiveness of domain-specific adaptations for LLMs in specialized fields.