Business Need

Radiology departments handle a high volume of CT imaging requests submitted as free-text clinical descriptions. Determining the correct imaging protocol for each order often requires manual review by radiologists or coordinators.

At Sahlgrenska Hospital, incorrect routing of CT orders led to protocol misclassification, repeated manual corrections, and delays in patient triage. The hospital required a reliable way to automatically interpret clinical order descriptions and route them to the appropriate imaging protocol while maintaining explainability and compliance with medical workflows.

Solution

AI-driven clinical text classification system that automatically maps CT order descriptions to the correct radiology protocol.

The solution combines medical language models and machine learning classification to interpret physician requests and accurately assign the appropriate protocol.

Key capabilities include:

• NLP-based analysis of free-text clinical CT orders

• Medical language embeddings optimized for domain-specific terminology

• Machine learning classification predicting the appropriate protocol from multiple possible categories

• Synthetic data generation pipeline to address class imbalance in medical datasets

• Seamless integration with hospital routing workflows for automated order handling

Results

• 98% classification accuracy on clinical test data

• Significant reduction in protocol misrouting

• Fewer manual corrections by radiologists

• Faster patient routing to the correct specialist

• Improved operational efficiency across radiology teams