A proof of concept study for machine learning application to stenosis detection.

Area of Science:

• Cardiovascular physiology
• Biomedical engineering
• Machine learning applications

Background:

• Arterial stenosis detection is crucial for cardiovascular health.
• Current diagnostic methods have limitations in accuracy and specificity.
• Hemodynamic data offers a potential avenue for improved stenosis assessment.

Purpose of the Study:

• To evaluate machine learning (ML) classifiers for predicting arterial stenosis.
• To assess the efficacy of ML in a three-vessel arterial system model.
• To compare ML performance against existing clinical methods.

Main Methods:

• Creation of a virtual patient database (VPD) using a 1D pulse wave propagation model.
• Training and testing of four ML classifiers (binary and multiclass) on hemodynamic data (pressure, flow-rate).
• Analysis of classifier performance using specificity, sensitivity, and area under the ROC curve.

Main Results:

• ML classifiers achieved specificities >80% and sensitivities of 50-75%.
• The best classifier achieved an AUC of 0.75, outperforming ~20 clinical methods.
• Key findings include the efficiency of using fewer measurements and identifying informative signals.

Conclusions:

• ML classifiers demonstrate moderate accuracy in detecting arterial stenosis.
• The developed method can identify the specific stenosed vessel, a significant advancement.
• Few hemodynamic measurements can yield high classification accuracy, suggesting clinical feasibility.