Coronary computed tomography angiographic detection of in-stent restenosis via deep learning reconstruction: a feasibility study

Objectives Evaluation of in-stent restenosis (ISR), especially for small stents, remains challenging during computed tomography (CT) angiography. We used deep learning reconstruction to quantify stent strut thickness and lumen vessel diameter at the stent and compared it with values obtained using conventional reconstruction strategies. Methods We examined 166 stents in 85 consecutive patients who underwent CT and invasive coronary angiography (ICA) within 3 months of each other from 2019-2021 after percutaneous coronary intervention with coronary stent placement. The presence of ISR was defined as percent diameter stenosis >= 50% on ICA. We compared a super-resolution deep learning reconstruction, Precise IQ Engine (PIQE), and a model-based iterative reconstruction, Forward projected model-based Iterative Reconstruction SoluTion (FIRST). All images were reconstructed using PIQE and FIRST and assessed by two blinded cardiovascular radiographers. Results PIQE had a larger full width at half maximum of the lumen and smaller strut than FIRST. The image quality score in PIQE was higher than that in FIRST (4.2 +/- 1.1 versus 2.7 +/- 1.2, p < 0.05). In addition, the specificity and accuracy of ISR detection were better in PIQE than in FIRST (p < 0.05 for both), with particularly pronounced differences for stent diameters < 3.0 mm. Conclusion PIQE provides superior image quality and diagnostic accuracy for ISR, even with stents measuring < 3.0 mm in diameter.

Automated summary

In this study, deep learning reconstruction was used to evaluate in-stent restenosis, showing superior image quality and diagnostic accuracy compared to conventional reconstruction strategies, especially for small stents.