The evolution of computing power has revolutionized the field of medical imaging, particularly in the realm of computed tomography (CT). One of the significant milestones in this evolution has been the introduction of iterative reconstruction methods, aimed at reducing image noise and radiation dose in CT scans. Initially, these methods were combined with filtered back projection to optimise computational costs. However, as computer performance has advanced, model-based iterative reconstruction (MBIR) has emerged as a superior alternative, particularly beneficial in coronary CT angiography due to its ability to enhance spatial resolution while mitigating image noise. Nonetheless, MBIR's Achilles heel remains its prolonged reconstruction time.


Model-Based Iterative Reconstruction (MBIR): Benefits and Limitations

In recent years, the advent of deep learning reconstruction (DLR) has marked another pivotal shift in CT image reconstruction. Distinct from MBIR, DLR offers the dual advantage of reducing image noise while significantly cutting down on reconstruction time. Super-resolution deep learning reconstruction (SR-DLR) takes this a step further by leveraging ultra-high-resolution CT data to augment spatial resolution. Several comparative studies, albeit with limitations, have showcased SR-DLR's potential superiority over other reconstruction methods, particularly in coronary CT angiography. A recent single-centre retrospective study delved into this comparative analysis, focusing on the objective and subjective image quality and diagnostic performance of MBIR versus SR-DLR in coronary artery stenosis detection via coronary CT angiography. The study encompassed 60 patients, with half undergoing scans reconstructed with MBIR and the other half with SR-DLR. Notably, the results indicated that SR-DLR outperformed MBIR in terms of reducing image noise and enhancing signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) at reduced tube current settings.


Diagnostic Performance: SR-DLR vs. MBIR

Subjectively, SR-DLR images received superior scores in terms of image quality, specifically in noise reduction and border conspicuity. Interestingly, while SR-DLR showcased notable improvements, both methods exhibited comparable diagnostic performance in identifying significant coronary stenosis. This finding contrasts with previous studies, which hinted at SR-DLR's potential to augment diagnostic accuracy and specificity, especially in segments with stents or severe calcification. The diagnostic prowess of coronary CT angiography using MBIR is well-established, boasting an accuracy exceeding 85%. This high diagnostic yield might explain why DLR, despite its advancements, doesn't necessarily outshine MBIR in overall coronary stenosis detection. However, there's potential for DLR to refine the accuracy of functional stenosis assessments, like CT fractional flow reserve, given that computational fluid dynamics' precision hinges on coronary structure reproducibility.


Radiation Dose Reduction: A Double-Edged Sword

One of the most significant advantages of novel reconstruction methods like DLR is the reduced radiation dose during CT angiography. MBIR has already paved the way for sub-millisievert radiation doses in coronary CT angiography. However, the study found that SR-DLR didn't achieve the expected reduction in tube current, primarily due to overlapping organs affecting tube current determination. Excessive radiation dose reduction risks negating the benefits of SR-DLR, as evidenced by previous studies comparing DLR and hybrid iterative reconstruction in abdominal CT scans. Despite its insights, the study had its limitations, including its single-centre nature, a relatively small sample size, and exclusions like patients with helical scans. Additionally, the study didn't explore the potential of DLR in diagnosing functional coronary stenosis or compare SR-DLR with normal-resolution DLR.


The study underscores SR-DLR's potential to elevate both objective and subjective image quality in coronary CT angiography while maintaining diagnostic performance on par with MBIR. As technology continues to evolve, further research with larger cohorts and diverse algorithms will undoubtedly elucidate the optimal balance between image quality and radiation dose, further cementing SR-DLR's role in advancing coronary CT angiography.


Source & Image Credit: Radiology Advances



Nobuo Tomizawa, Yui Nozaki, Hideyuki Sato et al. Super-resolution deep learning reconstruction to improve image quality of coronary CT angiography, Radiology Advances, 2024; umae001

Latest Articles

CT image reconstruction, Model-Based Iterative Reconstruction (MBIR), Deep Learning Reconstruction (DLR), coronary CT angiography, image quality, radiation dose reduction Explore the latest advancements in CT image reconstruction methods, including Model-Based Iterative Reconstruction (MBIR) and Deep Learning Reconstruction (DLR), and their impact on coronary CT angiography.