A deep learning algorithm may automate intracranial aneurysm detection on MR angiography with high diagnostic
Bio Joo1, Sung Soo Ahn2, Pyeong Ho Yoon3
1Department of Radiology, Research Institute of Radiological Science and Center for Clinical Image Data Science, Yonsei University College of Medicine, 50 Yonsei-ro, Seodaemun-gu, Seoul, 120-752, Korea.
European Radiology
|June 1, 2020
Summary
A new deep learning algorithm accurately detects intracranial aneurysms using MR angiography. This AI tool shows high diagnostic performance, validated across multiple institutions, suggesting potential for widespread clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neurology
Background:
- Intracranial aneurysms pose significant risks, necessitating accurate and efficient diagnostic methods.
- Time-of-flight MR angiography is a key imaging modality for detecting these vascular abnormalities.
Purpose of the Study:
- To develop and evaluate a deep learning algorithm for automated detection and localization of intracranial aneurysms.
- To assess the diagnostic performance of the algorithm using time-of-flight MR angiography data.
Main Methods:
- A retrospective, multicenter study utilizing MR images from radiological reports.
- A deep learning algorithm based on 3D ResNet architecture was trained on 468 examinations.
- Performance was evaluated on internal (120 aneurysm + 50 non-aneurysm) and external (56 aneurysm + 50 non-aneurysm) test sets.
Main Results:
- The algorithm achieved high diagnostic accuracy on both internal and external test sets.
- Internal test set: 87.1% sensitivity, 92.8% positive predictive value, 92.0% specificity.
- External test set: 85.7% sensitivity, 91.5% positive predictive value, 98.0% specificity.
Conclusions:
- A deep learning algorithm demonstrates high diagnostic performance for intracranial aneurysm detection.
- Validation with an external dataset from a different institution confirms the algorithm's robustness.
- The algorithm shows promise for reliable clinical application in real-world settings.
