MRI Compatibility: Automatic Brain Shunt Valve Recognition using Feature Engineering and Deep Convolutional Neural...

Area of Science:

• Medical Imaging
• Machine Learning
• Biomedical Engineering

Background:

• Cerebrospinal fluid shunt valves (CSF-SVs) are crucial for treating hydrocephalus.
• Accurate identification of CSF-SV models is necessary for MRI safety protocols.
• Current methods for identifying CSF-SVs can be time-consuming and prone to error.

Purpose of the Study:

• To develop and evaluate a machine learning method for distinguishing different models of CSF-SVs from clinical X-ray images.
• To assess the feasibility of using X-ray imaging and deep learning for an automated MRI safety system.
• To improve patient care by potentially reducing MRI delays or denials for individuals with implanted devices.

Main Methods:

• A retrospective study analyzed 416 skull X-rays containing CSF-SVs from common US brands.
• Four machine learning pipelines were compared: two using engineered features (LBP, HOG) and two using deep convolutional neural networks (CNNs).
• Performance was evaluated using accuracy, precision, recall, and F1-score, with confidence intervals determined by bootstrap procedures.

Main Results:

• The best-performing deep CNN method achieved 96% accuracy (95% CI: 94-98%) in identifying CSF-SV types.
• Deep CNN pipelines significantly outperformed engineered feature methods (95-96% vs. 56-64% mean accuracy).
• Stratified cross-validation was employed to ensure robust performance and prevent overfitting.

Conclusions:

• Machine learning, particularly deep convolutional neural networks, can effectively distinguish CSF-SV models from clinical X-rays.
• This automated approach represents a significant advancement towards developing a reliable MRI safety system for patients with CSF-SVs.
• The findings pave the way for enhanced patient safety and streamlined MRI procedures.