Automated Radiology-Arthroscopy Correlation of Knee Meniscal Tears Using Natural Language Processing Algorithms

Area of Science:

• Orthopedic Surgery
• Radiology
• Computer Science

Background:

• Accurate correlation between radiology and arthroscopy is crucial for diagnosing meniscal tears.
• Manual correlation is time-consuming and prone to errors.
• Automating this process can improve efficiency and accuracy.

Purpose of the Study:

• To train and apply natural language processing (NLP) algorithms for automated radiology-arthroscopy correlation of meniscal tears.
• To assess the performance of NLP models in detecting meniscal tears from MRI and arthroscopy reports.
• To evaluate the ability of NLP to identify discrepancies between radiology and arthroscopy findings.

Main Methods:

• Trained supervised machine learning models (logistic regression, SVM, random forest) on 3593 manually annotated knee MRI reports to detect meniscal tears.
• Evaluated NLP model performance using cross-validation and a separate test set (108 MRIs with subsequent arthroscopy).
• Assessed radiology-arthroscopy agreement by comparing ensembled NLP-extracted findings with manually annotated findings.

Main Results:

• NLP models achieved high cross-validation performance for meniscal tear detection on MRI reports (medial meniscus F1: 0.93-0.94, lateral meniscus F1: 0.86-0.88).
• When evaluated on arthroscopy reports, NLP performance was similar, with ensembling improving results (medial meniscus F1: 0.97, lateral meniscus F1: 0.99).
• Ensembled NLP models detected radiology-arthroscopy mismatches with 79% sensitivity and 87% specificity.

Conclusions:

• Automated radiology-arthroscopy correlation for knee meniscal tears using NLP is feasible.
• This approach holds potential for enhancing medical education and quality improvement in orthopedic surgery.
• NLP offers a valuable tool for identifying discrepancies in meniscal tear reporting.