Can Natural Language Processing Improve Adnexal Torsion Predictions?

Area of Science:

• Gynecology
• Medical Informatics
• Machine Learning

Background:

• Adnexal torsion is a gynecological emergency requiring prompt diagnosis.
• Accurate prediction of adnexal torsion is challenging, often leading to delays or unnecessary surgical interventions.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) and natural language processing (NLP) based decision support tool for predicting adnexal torsion.
• To augment clinicians' diagnostic capabilities for suspected adnexal torsion.

Main Methods:

• A retrospective cohort study was conducted using data from 433 women undergoing surgery for suspected adnexal torsion.
• Electronic medical records were analyzed, with NLP extracting data from unstructured text.
• A CatBoost classifier model was trained to predict adnexal torsion based on clinical and sonographic data.

Main Results:

• The ML/NLP model achieved an 84% prediction accuracy for adnexal torsion, with a 95% recall for torsion cases.
• Key predictors identified included patient age, and differences in ovarian size.
• The model demonstrated 77% precision for identifying cases without torsion, though recall for this group was 45%.

Conclusions:

• Machine learning and NLP offer a feasible approach to creating a decision support tool for adnexal torsion diagnosis.
• The developed tool shows potential to improve diagnostic accuracy and reduce the rate of unnecessary laparoscopies for suspected adnexal torsion.