Abbreviation Expansion Using a Single Word as Context.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Decision Support

Background:

• Medical abbreviations are frequently ambiguous, posing challenges for accurate clinical interpretation and decision-making.
• Automated methods are needed to resolve abbreviation ambiguity in large medical text corpora.

Purpose of the Study:

• To develop and evaluate a simple Word Sense Disambiguation (WSD) method for expanding ambiguous medical abbreviations.
• To assess the performance of an Odds Ratio (OR)-based approach against established machine learning models.

Main Methods:

• A knowledge dictionary was constructed using the Odds Ratio (OR) of surrounding word tokens from a large medical corpus.
• Abbreviation expansion was performed by identifying the most informative word from the generated dictionary.
• The OR-based method was evaluated on ambiguous abbreviations from the NCBI Disease Corpus test set.

Main Results:

• The OR-based abbreviation disambiguation method achieved an accuracy of 0.82.
• Performance closely approached that of Support Vector Machine (SVM) at 0.87 and BERT models at 0.89.
• The method demonstrated effectiveness despite its inherent simplicity.

Conclusions:

• The proposed simple Odds Ratio (OR)-based method is an effective and interpretable approach for medical abbreviation disambiguation.
• This technique shows potential as a precise alternative for clinical text analysis.
• Further development could enhance its utility in clinical decision support systems.