Deep Learning-Based Automated Coronary Plaque Quantification with Ultra-high Resolution Photon-Counting Detector CT...

To evaluate the impact of varying spatial resolutions on image quality, and deep-learning (DL) based quantification of coronary stenosis and plaque volumes using photon-counting detector CT (PCD-CT).

In this prospective single-center study, 70 patients who were suggestive of coronary artery disease (CAD) and had undergone UHR-CCTA on the PCD-CT system were included. The CCTA datasets were reconstructed with slice thicknesses of standard resolution (SR) 0.6 mm, high spatial resolution (HR) 0.4 mm, and ultra-high resolution (UHR) 0.2 mm. Two radiologists assessed subjective image quality, while a DL-based tool automatically quantified the degree of stenosis and plaque volumes across the three resolutions. The extent of stenosis and plaque volumes were compared between UHR, HR, and SR using the Friedman test with post hoc testing, which had Bonferroni correction (0.05/3 = .017).

Subjective assessment revealed that higher spatial resolution enhanced vessel sharpness, and stronger calcified and partially calcified plaque diagnosis confidence, but concurrently increased perceived image noise (both P < 0.05). The DL-based stenosis evaluation of 139 coronary artery plaques revealed a decreasing trend of stenosis diameter with improving spatial resolution for calcified and partially calcified plaques, particularly for calcified plaques. (UHR, HR, and SR, respectively: 15% [IQR, 10-20%], 19% [IQR, 10-20.5%], and 20% [IQR, 11-21%]; P = 0.011), whereas noncalcified plaques did not show evidence of a difference (P = 0.292). The median plaque volume exhibits a downward trend as the resolution increases (for overall plaque volume, 29.05 mm³[IQR, 13.42 mm³, 58.86 mm³], 31.9 mm³[IQR, 16.97 mm³, 66.82 mm³],34.21 mm³ [IQR, 17.13 mm³, 67.9 mm³], P < 0.001, respectively).

Although UHR PCD-CT increases perceived image noise, its superior spatial resolution enhances vessel sharpness and significantly reduces the overestimation of DL-based quantitative plaque volume and stenosis severity in calcified and partially calcified lesions. Recognizing this resolution-dependent variability is essential for reliable CAD risk stratification.