Ascle: A Python Natural Language Processing Toolkit for Medical Text Generation.

Area of Science:

• Biomedical Informatics
• Natural Language Processing

Background:

• Medical text generation requires specialized tools for researchers and clinicians.
• Existing natural language processing (NLP) solutions often demand significant programming expertise.
• There is a need for integrated, user-friendly platforms for advanced medical NLP tasks.

Purpose of the Study:

• Introduce Ascle, a novel NLP toolkit for medical text generation.
• Provide an all-in-one solution with advanced generative functions and essential NLP capabilities.
• Evaluate the performance of fine-tuned language models on medical text tasks.

Main Methods:

• Fine-tuned 32 domain-specific language models.
• Evaluated models on 24 established benchmarks for text generation tasks.
• Conducted manual reviews with clinicians for question-answering evaluation, assessing Readability, Relevancy, Accuracy, and Completeness.

Main Results:

• Fine-tuned models demonstrated consistent improvements in text generation.
• Achieved a 20.27 increase in BLEU score for machine translation.
• Manual reviews for answer generation yielded high average scores: 4.95 (Readability), 4.43 (Relevancy), 3.9 (Accuracy), and 3.31 (Completeness).

Conclusions:

• Ascle is a user-friendly NLP toolkit for medical text generation.
• The toolkit supports advanced functions like question-answering, summarization, simplification, and translation.
• Ascle, its models, and data are publicly available to facilitate research and development.