Classification of anatomic patterns of peripheral artery disease with automated machine learning (AutoML)

Area of Science:

• Vascular Medicine
• Artificial Intelligence
• Machine Learning

Background:

• Peripheral artery disease (PAD) diagnosis relies on anatomical pattern classification.
• Current methods may be subject to human variability and fatigue.
• Automated machine learning (AutoML) offers a potential solution for objective classification.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) model using AutoML for classifying peripheral artery disease (PAD) anatomical patterns.
• To assess the model's performance in grading femoropopliteal disease using the Global Limb Anatomic Staging System (GLASS).

Main Methods:

• Collected and labeled 323 lower limb angiograms using the Global Limb Anatomic Staging System (GLASS).
• Trained an AutoML model on the Vertex AI platform for multi-label classification of GLASS grades.
• Evaluated the model on 25 test angiograms and performed incremental training for performance improvement.

Main Results:

• The AI model achieved an initial average precision of 0.77, correctly classifying GLASS grades in 100% of test cases.
• Agreement with expert classification was high (Kappa = 0.85), with best performance in grades 0 and 4.
• Incremental training improved average precision by 11% to 0.86, demonstrating adaptability.

Conclusions:

• AutoML shows significant potential for accurate and objective classification of PAD anatomical patterns.
• This technology can enhance outcome prediction and standardize revascularization strategies in vascular medicine.
• Further research and financial support are warranted to fully realize the benefits of AutoML in PAD management.