Angiography-based coronary flow reserve: The feasibility of automatic computation by artificial intelligence

Area of Science:

• Cardiovascular Imaging
• Medical Technology
• Computational Cardiology

Background:

• Coronary flow reserve (CFR) is a significant prognostic indicator in patients with coronary artery disease.
• Current CFR measurement methods are complex and lack widespread automated solutions.
• There is a need for accessible tools to assess coronary physiology.

Purpose of the Study:

• To develop and assess the feasibility of an automated method for CFR computation using coronary angiography.
• To evaluate the performance of a convolutional neural network (CNN) for coronary artery segmentation.
• To compare automated CFR (CFRauto) with manual CFR (CFRmanual) calculations.

Main Methods:

• A CNN model was trained on annotated coronary angiograms for automatic segmentation of coronary arteries.
• The CNN model was integrated into a software prototype for automated CFR calculation (CFRauto).
• CFRauto was applied to patient angiograms and compared with manual frame-count-based CFR (CFRmanual).

Main Results:

• The CNN model demonstrated good performance in segmenting coronary arteries.
• CFRauto was successfully computed in 91.3% of analyzed vessels.
• A moderate correlation (r = 0.51) was found between CFRauto and CFRmanual, with a mean difference of 0.12.

Conclusions:

• Automated CFR computation from coronary angiography is feasible.
• This automated method has the potential to enhance the clinical application of coronary physiology.
• Wider adoption of coronary physiology in catheterization labs for microcirculatory function assessment is facilitated.