A Machine Learning System to Automate Body Computed Tomography Protocoling.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Radiology protocol selection is critical for patient health and healthcare costs.
• Current protocol selection is often inefficient and time-consuming for radiologists.
• Suboptimal protocol selection can lead to delayed treatment and increased healthcare expenses.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) system for accurate radiology protocol prediction.
• To improve the efficiency of the radiology workflow through automated protocol selection.
• To leverage electronic medical record (EMR) data for predicting optimal imaging protocols.

Main Methods:

• Developed an ensemble ML system using three decision tree (DT)-based techniques.
• Trained models on the 15 most common body computed tomography (CT) abdomen protocols.
• System designed to provide the top three most probable protocol predictions for radiologist review.

Main Results:

• The ensemble ML classifier achieved an F1 score of approximately 83% in 5-fold cross-validation.
• The system demonstrated a high performance for the top three predictions with an F1 score of 95.5%.
• The ensemble model outperformed individual DT-based models, which had mean F1 scores around 80% and individual prediction F1 scores from 87.6% to 92.9%.

Conclusions:

• Machine learning techniques can accurately predict radiology protocols from EMR data.
• The developed ML system can serve as a clinical decision support tool.
• This approach has the potential to significantly improve radiologist efficiency and optimize patient care.