Deep Learning-Assisted Diagnosis of Cerebral Aneurysms Using the HeadXNet Model.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning offers potential for improving medical image interpretation.
• Automated segmentation can reduce diagnosis time.
• Few studies have explored AI's role in augmenting clinician performance for intracranial aneurysm detection.

Purpose of the Study:

• To develop and apply a neural network segmentation model (HeadXNet) for precise intracranial aneurysm detection on head CT angiography (CTA).
• To evaluate if AI-generated segmentations augment clinicians' diagnostic performance for intracranial aneurysms.

Main Methods:

• A 3D convolutional neural network was trained on 611 head CTA scans.
• Segmentation outputs were provided to 8 clinicians for diagnosing aneurysms on a separate test set of 115 scans.
• Clinician performance was compared with and without AI augmentation in a crossover design.

Main Results:

• AI augmentation significantly improved clinician sensitivity (0.059 increase, P=.01) and accuracy (0.038 increase, P=.02).
• Interrater agreement also improved with AI assistance (Fleiss κ increased by 0.060, P=.05).
• No significant changes were observed in specificity or time to diagnosis.

Conclusions:

• The developed deep learning model accurately detected intracranial aneurysms on CTA.
• AI-assisted diagnosis shows potential to augment clinician performance, leading to more dependable and accurate predictions.
• Integration of such models may optimize patient care through improved diagnostic accuracy.