Deep Learning Detection of Aneurysm Clips for Magnetic Resonance Imaging Safety

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Patient Safety

Background:

• Metal implants, such as aneurysm clips, pose significant risks during MRI scans.
• Current methods for identifying patients with aneurysm clips before MRI are manual and can be error-prone.
• Automated detection of aneurysm clips is needed to enhance patient safety.

Purpose of the Study:

• To evaluate the accuracy of a machine learning model in classifying the presence or absence of aneurysm clips on CT images.
• To develop an automated system for flagging aneurysm clips prior to MRI appointments.

Main Methods:

• A dataset of 280 CT head scans (140 with and 140 without aneurysm clips) was used.
• A pre-trained image classification neural network was retrained to classify CT localizer images.
• The model underwent fivefold cross-validation and testing on a holdout set.
• SHapley Additive exPlanations (SHAP) were used to interpret model predictions.

Main Results:

• The machine learning model achieved a mean sensitivity of 100% and a mean accuracy of 82% in detecting aneurysm clips on CT localizer images.
• SHAP analysis confirmed that the model focused on relevant regions of interest.
• Models trained on 3D CT head scans did not outperform the localizer models.

Conclusions:

• Machine learning, specifically computer vision image classification, can accurately detect aneurysm clips on CT scans.
• This automated approach has the potential to significantly improve patient safety by enhancing pre-MRI screening processes.