Negation and uncertainty detection in clinical texts written in Spanish: a deep learning-based approach

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Accurate medical text mining requires detecting negation and uncertainty to prevent misinterpretation of clinical events.
• Existing methods for Spanish clinical texts primarily focus on negation, neglecting uncertainty detection.
• A gap exists in robust methods for identifying both negation and uncertainty in Spanish clinical narratives.

Purpose of the Study:

• To propose and evaluate a deep learning-based approach for simultaneous negation and uncertainty detection in Spanish clinical texts.
• To compare the effectiveness of Bidirectional Long-Short Term Memory with a Conditional Random Field layer (BiLSTM-CRF) and Bidirectional Encoder Representation for Transformers (BERT) for this task.
• To validate the approach on public corpora and a real-world clinical dataset.

Main Methods:

• Implementation of two deep learning models: BiLSTM-CRF and BERT.
• Evaluation using the NUBES and IULA Spanish language corpora.
• Validation on an annotated clinical notes dataset from cancer patients.

Main Results:

• Achieved an F-score of 92% for negation scope recognition and 80% for uncertainty scope recognition.
• Demonstrated the feasibility of deep learning for detecting both negation and uncertainty in Spanish clinical texts.
• Showcased improved performance in scope recognition compared to existing biomedical domain proposals.

Conclusions:

• Deep learning models, specifically BiLSTM-CRF and BERT, are effective for detecting negation and uncertainty in Spanish clinical texts.
• The proposed approach enhances the reliability of medical text mining applications for Spanish clinical data.
• This work addresses a significant limitation in current Spanish clinical NLP by handling both negation and uncertainty detection.