BoneScore: A natural language processing algorithm to extract bone mineral density data from DXA scans.

Area of Science:

• Medical Informatics
• Computational Linguistics
• Radiology

Background:

• Dual-energy X-ray absorptiometry (DXA) scans are crucial for diagnosing osteoporosis.
• Extracting specific T-score data from DXA reports can be labor-intensive and prone to error.
• Automated methods are needed to efficiently and accurately process this clinical information.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm named BoneScore.
• To accurately detect and extract femoral neck T-score information from DXA scan reports.
• To assess the clinical validity of the extracted T-score data.

Main Methods:

• A rule-based NLP algorithm was developed, iteratively creating regular expressions.
• The algorithm, BoneScore, processed 889 text snippets from DXA reports.
• Accuracy was optimized by testing word window lengths around the key term 'femoral'; clinical validation involved regression analysis.

Main Results:

• BoneScore achieved 98% accuracy in detecting T-score information using 20 regular expressions and a 50-word window.
• Extracted T-scores demonstrated associations with established osteoporosis risk factors in regression models.
• The algorithm successfully extracted T-score values with high precision.

Conclusions:

• BoneScore effectively uses regular expressions to accurately extract bone mineral density T-scores from DXA reports.
• The algorithm demonstrates clinical face validity, supporting its utility in osteoporosis assessment.
• This NLP approach offers a reliable method for processing critical T-score data.