Machine Learning and Deep Learning Models for Automated Protocoling of Emergency Brain MRI Using Text from Clinical...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Natural Language Processing (NLP) for Healthcare
• Radiology Workflow Optimization

Background:

• Automated protocoling of emergency brain MRI scans is crucial for efficient patient care.
• Clinical referral text contains vital information for determining appropriate imaging protocols.
• Existing manual protocoling methods can be time-consuming and prone to variability.

Purpose of the Study:

• To develop and evaluate machine learning (ML) and deep learning (DL) models for automated emergency brain MRI protocoling.
• To predict the correct MRI imaging protocol and the necessity of contrast agent administration based on referral text.
• To compare the performance of ML and DL models against human expert and non-expert performance.

Main Methods:

• Retrospective analysis of 1953 emergency brain MRI referrals.
• Development of three ML models (naive Bayes, SVM, XGBoost) and two DL models (BERT, GPT-3.5).
• Models trained on varying dataset sizes (100%, 50%, 50% + augmented data) and evaluated on a test set.

Main Results:

• GPT-3.5 models achieved the highest accuracy: 84% for protocol and 91% for contrast agent.
• BERT models showed strong performance with 78% protocol accuracy and 89% contrast agent accuracy.
• ML models, particularly XGBoost and SVM, performed competitively, with accuracies up to 78% for protocol and 88% for contrast agent.

Conclusions:

• ML and DL models demonstrate high efficacy in automating emergency brain MRI protocoling from clinical referral text.
• The performance of these AI models approaches that of human neuroradiologists and non-experts.
• Automated protocoling using NLP offers a promising avenue for enhancing efficiency and consistency in emergency radiology.