AI Shows High Accuracy in CT, MRI Protocoling

Meta-analysis finds similar protocoling performance across machine learning, BERT-based models, and large language models.

1

AI systems can achieve approximately 85% accuracy in assigning CT and MRI examination protocols, based on a meta-analysis of 23 studies.
2

Transformer-based models showed the highest accuracy at 87%, while traditional machine learning models achieved 83% accuracy.
3

The study reviewed 57 performance results from 30 distinct models, with training datasets ranging from 1,235 to 559,305 cases.
4

Common protocoling errors stemmed from ambiguous requisition text and data imbalance, affecting model performance on rare categories.
5

Hybrid AI-radiologist systems may optimize workflow by automating routine protocols while allowing radiologist review for uncertain cases.

1