Deep learning-based classification of lower extremity arterial stenosis in computed tomography angiography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Peripheral artery disease (PAD) affects lower extremity arteries.
• Accurate assessment of arterial stenosis is crucial for PAD management.
• Computed tomography angiography (CTA) is a key imaging modality for PAD.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying lower extremity artery stenosis.
• To assist radiologists in interpreting CTA scans for PAD patients.
• To quantify arterial plaque-induced stenosis in lower extremity arteries.

Main Methods:

• A retrospective study of 265 patients with lower extremity CTA scans.
• A parallel efficient network (p-EffNet) was trained on 17050 arterial images.
• Digital subtraction angiography served as the reference standard for stenosis classification.

Main Results:

• The p-EffNet achieved high accuracy and AUC for both above-knee (91.5% accuracy, 0.987 AUC) and below-knee arteries (90.9% accuracy, 0.981 AUC).
• Model performance showed comparable accuracy and specificity to radiologists.
• The model demonstrated lower sensitivity compared to human readers for both artery segments.

Conclusions:

• The p-EffNet exhibits strong diagnostic performance for lower extremity artery stenosis.
• This deep learning model has the potential to reduce radiologist workload.
• The tool may help identify overlooked or misjudged arterial plaques in PAD patients.