Super Resolution Deep Learning Reconstruction for Cardiac CT: upscale your game!

Congress:

ECR 2023

Poster Number:

C-10591

Type:

Educational Exhibit

Keywords:

Cardiac, Computer applications, CT, CT-Angiography, CT-High Resolution, eLearning, Education and training

Authors:

M. Ohana, F. Tatsugami, A. Labani, W. Fukumoto, A. Taniguchi, K. Haioun, K. Awai, C. Roy

DOI:

10.26044/ecr2023/C-10591

DOI-Link:

https://dx.doi.org/10.26044/ecr2023/C-10591

Learning objectives

Understand Deep Learning Reconstruction (DLR) and Super Resolution Deep Learning Reconstruction (SR-DLR) application in cardiac CT Apprehend the usage of SR-DLR in routine clinical practice Assimilate clinical examples of SR-DLR Be aware of current limitations and future directions

Background

Ultra-High Resolution (UHR) coronary-CTA (CCTA) acquisitions are available since 2017 (Aquilion Precision), with : 0.25mm detector size 160 rows 0.4 x 0.5 mm focal spot 1024 and 2048 matrix sizes Deep Learning Reconstructions are feasible Publications such as [1] have demonstrated significant advances in CCTA diagnostic performances with UHR-CT in patients with heavily calcified arteries (Figure 1) and/or stents. [Fig 1] However, UHR CT is – in 2023 – still largely under-deployed [2], due to High cost Inadequate PACS infrastructure Some technical disadvantages in the...

Findings and procedure details

What are the expected benefits of SR-DLR in cardiac CT? 1/ An increased perceived Image Quality In cardiac CT – including coronary CTA, valve CT and myocardium analysis, perceived image quality is higher with SR-DLR when compared to DLR and Hybrid IR. This increased image quality appears to be the resultant of a combination of factors. A/ Significant noise reduction (Figure 4) - Even compared with DLR, with about an additional 20% reduction - This reduction is present in the whole CT slice – even...

Conclusion

SR-DLR in - routine - cardiac CT in 2023 Is an AI algorithm designed to enhance/increase CT image quality Is dedicated – for now – to cardiac CT Is usable without impact on workflow nor dosimetry. SR-DLR does increase the perceived IQ of coronary CTA. SR-DLR could carry a potential diagnostic impact, with possible better delineation of minimal lesions possible better assessment of highly calcified vessels possible better plaque quantification.

Personal information and conflict of interest

M. Ohana: Speaker: Canon Medical Systems Europe Speaker: Boehringer Ingelheim F. Tatsugami: Nothing to disclose A. Labani: Nothing to disclose W. Fukumoto: Nothing to disclose A. Taniguchi: Employee: Canon Medical Systems Japan K. Haioun: Employee: Canon Medical Systems Japan K. Awai: Nothing to disclose C. Roy: Nothing to disclose

References

Latina, J., Shabani, M., Kapoor, K., Whelton, S. P., Trost, J. C., Sesso, J., ... & Arbab-Zadeh, A. (2021). Ultra-high-resolution coronary CT angiography for assessment of patients with severe coronary artery calcification: initial experience.Radiology: Cardiothoracic Imaging Schuijf, J. D., Lima, J. A., Boedeker, K. L., Takagi, H., Tanaka, R., Yoshioka, K., & Arbab-Zadeh, A. (2022). CT imaging with ultra-high-resolution: Opportunities for cardiovascular imaging in clinical practice.Journal of cardiovascular computed tomography Lee, T. C., Zhou, J., Yu, Z., & Cai, L. (2022, April). Deep learning enabled...

Fig 1: Example of UHR-CT advantage over conventional CT in coronary CTA: patient with...

Fig 2: Training of SR-DLR in the factory, with a database consisting of paired images...

Fig 3: Onsite application of SR-DLR to enhance conventional CT images.

Fig 4: Significant noise reduction with SR-DLR - inside or outside the heart.

Fig 5: Increased sharpness of all contours - for regular or subtle lesions.

Fig 6: Increased sharpness for curved MPRs - consequence of a 3D CNN.

Fig 7: Better delineation of small structures - small arterial branches shown here.

Fig 8: Increased conspicuity of calcifications - useful for faint or low density...

Fig 9: Total number of abnormal coronary segments in patients with no to minimal CAD...

Fig 10: Moderate increase of blooming artifacts with SR-DLR, if adjacent to dense or...

Fig 11: Advantage of a 1024x1024 matrix size for plaque delineation and lesion analysis...

Fig 12: Advantage of a 1024x1024 matrix size for prosthetic valve delineation (work in...