A field theoretical approach to medical natural language processing

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Medical Informatics

Background:

• Parsing medical free text reports is challenging due to complex linguistic structures.
• Existing methods may not fully capture the nuanced dependencies within clinical narratives.

Purpose of the Study:

• To develop and evaluate a novel parser for medical free text reports.
• To leverage a chemistry/physics-inspired field theory for analyzing sentence-level word dependencies.

Main Methods:

• Developed a parser based on a "field theory" model for word-word dependencies.
• Utilized psycholinguistics principles to guide the transition from linguistic to physical models.
• Calibrated the model using n-gram experiments and trained/tested on a corpus of radiology reports.

Main Results:

• The parser generates a dependency tree representing the global minimum energy state of words in a sentence.
• Preliminary performance on radiology reports achieved 84.9% link recall and 89.9% precision.

Conclusions:

• The proposed field theory approach offers a promising method for parsing complex medical text.
• The parser demonstrates effective identification of word dependencies in clinical narratives.