Development of a convolutional neural network to detect abdominal aortic aneurysms

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Cardiovascular disease screening

Background:

• Infrarenal abdominal aortic aneurysms (AAAs) are a significant vascular condition.
• Early detection of AAAs is crucial for effective management and prevention of rupture.
• Computed tomography angiography (CTA) is a key imaging modality for AAA diagnosis.

Purpose of the Study:

• To develop and train a foundational convolutional neural network (CNN) for automated screening of CTA scans.
• To identify infrarenal abdominal aortic aneurysms (AAAs) using artificial intelligence.
• To establish a proof-of-concept for AI-driven predictive modeling in AAA management.

Main Methods:

• Retrospective analysis of 400 abdominopelvic CTA scans (200 AAA, 200 control).
• Training a CNN using transfer learning (VGG-16, ImageNet) for binary classification.
• Model validation and testing using 60% training, 10% validation, and 30% testing datasets.

Main Results:

• The custom CNN model achieved 99.1% accuracy and an area under the receiver operating characteristic curve of 0.99.
• Gradient-weighted class activation mapping identified misjudgments primarily in small aneurysms (<3.3 cm) with mural thrombus.
• The model demonstrated high accuracy in detecting infrarenal abdominal aortic aneurysms (AAAs) on CTA scans.

Conclusions:

• The developed CNN model accurately screens CTA scans for infrarenal AAAs.
• This AI model serves as a foundation for future predictive modeling and AI applications in vascular disease.
• The findings support the potential of AI in enhancing AAA screening and diagnosis.