Natural Language Processing of Radiology Reports in Patients With Hepatocellular Carcinoma to Predict Radiology...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Radiology Resource Management

Background:

• Radiology services are in high demand, facing challenges in predicting patient needs due to uncertain disease prognoses.
• Accurate forecasting of radiology resource utilization is crucial for efficient healthcare management.

Purpose of the Study:

• To investigate the efficacy of Natural Language Processing (NLP) in predicting downstream radiology resource utilization.
• To assess NLP's potential for managing patients undergoing surveillance for hepatocellular carcinoma (HCC).

Main Methods:

• Utilized open-source NLP and machine learning (logistic regression, SVM, random forest) on radiology reports from HCC surveillance CT exams (2010-2017).
• Employed bag-of-words and TF-IDF (term frequency-inverse document frequency) for text representation.
• Evaluated model performance using accuracy, sensitivity, specificity, and AUC.

Main Results:

• TF-IDF feature extraction generally outperformed bag-of-words models.
• The TF-IDF + Support Vector Machine (SVM) model achieved the highest performance: 92% accuracy, 83% sensitivity, 96% specificity, and an AUC of 0.971.

Conclusions:

• NLP-based models demonstrate high accuracy in predicting radiology resource utilization from HCC surveillance reports.
• This predictive capability holds significant potential for improving healthcare decision-making, cost reduction, and expanding access to care.