Artificial intelligence in healthcare text processing: a review applied to named entity recognition

Area of Science:

• Natural Language Processing (NLP)
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Traditional methods (rule-based, word embeddings, sequence tagging) struggle with medical text complexity, variability, and data requirements.
• Named Entity Recognition (NER) is crucial for analyzing healthcare texts, identifying medical conditions and drug names.
• Effective NER supports information retrieval, personalized medicine, and clinical decision support.

Purpose of the Study:

• To systematically map and evaluate advanced language models, particularly transformer-based models like BERT, for medical NER.
• To compare the performance of transformer models against traditional and hybrid approaches in processing complex medical texts.
• To analyze the global contributions to this research area and discuss implications for healthcare systems.

Main Methods:

• Conducted a systematic mapping review focusing on transformer-based language models (e.g., BERT) for medical NER.
• Analyzed the models' ability to capture semantic dependencies and linguistic nuances in medical texts.
• Compared performance metrics (e.g., F1 scores) with traditional methods like CNNs and RNNs.

Main Results:

• Transformer-based models, especially BERT and ClinicalBERT, consistently achieve high performance, with F1 scores often exceeding 97%.
• These advanced models significantly outperform traditional and hybrid methods in medical NER tasks.
• China and the United States are leading contributors to research in this domain.

Conclusions:

• Transformer-based models represent a significant advancement in medical NER, offering superior accuracy and flexibility.
• The findings highlight the potential for integrating advanced NER technologies into healthcare systems, such as Brazil's SUS.
• Continuous updates on NLP advancements are vital for maximizing the relevance and application of NER in healthcare.