Deep learning models for triaging hospital head MRI examinations

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Increasing demand for head MRI scans and radiologist shortages cause reporting delays.
• Delayed MRI reports negatively impact patient outcomes and increase healthcare costs.
• Computer vision models offer potential to expedite head MRI interpretation.

Purpose of the Study:

• To develop and evaluate a deep learning framework for detecting clinically-relevant abnormalities in head MRI scans.
• To address the bottleneck of limited, labelled datasets in developing AI for neuroradiology.
• To assess the potential of AI in reducing head MRI reporting times for clinical triage.

Main Methods:

• Developed a convolutional neural network (CNN) framework for abnormality detection in head MRI.
• Utilized a Transformer-based classifier to generate a large labelled dataset (70,206 examinations).
• Trained and validated models on hospital-grade axial T2-weighted and diffusion-weighted MRI scans.

Main Results:

• Achieved fast (< 5s) and accurate (AUC > 0.9) classification of MRI abnormalities.
• Demonstrated good generalisability across two UK hospital networks (ΔAUC ≤ 0.02).
• Simulations showed potential to reduce mean reporting time for abnormal scans by up to 50%.

Conclusions:

• The deep learning framework shows feasibility for clinical triage of head MRI scans.
• AI-powered abnormality detection can significantly expedite the reporting process.
• This approach can help optimize radiologist resource allocation and improve patient care pathways.