The present study demonstrated that DLIR significantly improved the image quality of MDE-CT compared with conventional HIR. Quantitative metrics, including image noise, CNR, and SNR, were all significantly improved with DLIR. Both CNR and SNR increased by approximately 27%, while image noise decreased by over 20% on average. Qualitative assessments by two experienced radiologists also confirmed these findings, showing significantly higher image quality scores and an increased proportion of diagnostically suitable images with DLIR. These results suggest that DLIR can enhance the diagnostic performance of MDE-CT by improving the visualization of delayed myocardial enhancement.

MDE-CT has emerged as a valuable alternative to LGE-CMR for detecting and characterizing myocardial fibrosis and scar, particularly in patients with contraindications to CMR such as implanted devices, claustrophobia, or patients on dialysis. Several studies have shown that MDE-CT can identify myocardial infarction and non-ischemic fibrosis with diagnostic accuracy comparable to LGE-CMR [6, 15, 16]. Furthermore, MDE-CT offers rapid image acquisition and broader availability, making it a practical option in emergency settings or in institutions without ready access to CMR. Its clinical utility has been demonstrated in assessing myocardial viability, guiding therapeutic decisions, and predicting outcomes in both ischemic and non-ischemic cardiomyopathies [15, 17,18,19,20,21,22].

However, MDE-CT is subject to several image quality–limiting factors. Myocardial delayed enhancement is typically subtle in contrast, making it highly susceptible to degradation from image noise, motion artifacts, beam-hardening, and partial volume effects. These challenges are further compounded by the fact that MDE-CT is frequently performed in the same session as coronary CT angiography. To minimize cumulative radiation exposure, MDE-CT protocols often employ low-dose acquisition settings. In addition, low-energy imaging techniques, such as low tube voltage acquisition, are frequently adopted in MDE-CT to further enhance iodine contrast [14]. While such protocols are advantageous for reducing radiation dose and enhancing iodine contrast, they also increase the risk of poor signal-to-noise ratio, potentially limiting diagnostic reliability. These inherent limitations underscore the need for advanced reconstruction techniques that can improve the image quality of MDE-CT. In this context, DLIR may offer a promising solution. By leveraging deep convolutional neural networks trained on high-quality image datasets, DLIR can suppress image noise more effectively than conventional iterative techniques while preserving fine anatomical details and image texture [23, 24]. A recent study has applied deep learning–based reconstruction to MDE-CT and demonstrated improved image quality and inter-observer reproducibility compared with conventional reconstruction methods [25]. Our study showed that DLIR significantly improved image quality metrics—reducing noise by over 20% and increasing both CNR and SNR by approximately 27%. Subjective evaluations by two radiologists confirmed these findings, with higher image quality scores and a greater proportion of diagnostically suitable images in DLIR reconstructions. The increased proportion of diagnostically suitable images observed with DLIR may help reduce inter-reader variability and thereby improve the consistency of diagnostic interpretation in clinical settings. Although the impact on downstream clinical decision-making requires further investigation, the enhanced image quality achieved with DLIR supports its value in strengthening the technical robustness of MDE-CT. These findings underscore the potential of DLIR as a meaningful component in the ongoing optimization of cardiac CT protocols.

The substantial improvement in image quality achieved by DLIR may allow for further radiation dose reduction in MDE-CT protocols. However, the extent to which DLIR can contribute to dose reduction remains uncertain based on the present study. In particular, because the current MDE-CT protocol incorporates image averaging of four separate acquisitions to improve image quality, future investigations should explore whether suitable diagnostic quality can be achieved with fewer phases when using DLIR. In addition, further studies with larger and more diverse patient populations are needed to determine optimal acquisition strategies and clarify the clinical scenarios in which DLIR-driven dose reduction would be most beneficial, especially for younger patients or those requiring repeated imaging.

Several limitations of this study should be acknowledged. First, this was a retrospective study conducted at a single institution, which may introduce selection bias and limit the generalizability of the findings. Second, although the TrueFidelity DLIR algorithm offers three selectable noise reduction strength levels (low, medium, and high), only the highest strength level was evaluated in this study. Third, ASiR-V was used with a fixed 50% blending factor, which is a commonly adopted setting in routine clinical practice. However, variations in the blending ratio could affect image quality, and different settings may yield different comparative results. In this study, we aimed to evaluate the noise reduction potential of the latest DLIR method against ASiR-V under standard clinical parameters. Fourth, this study did not include an analysis of noise texture, which may affect the subjective perception of image quality. Fifth, although our results demonstrate that DLIR improves both quantitative and qualitative image quality, the study did not evaluate diagnostic performance metrics such as sensitivity, specificity, or accuracy for detecting myocardial scar or fibrosis. Consequently, whether the observed improvements in image quality translate into enhanced diagnostic capability remains uncertain and warrants further investigation in future studies. Finally, patients without visible MDE on CT were excluded, which may have introduced selection bias and precluded assessment of false-positive findings.

In conclusion, this study demonstrates that DLIR significantly improves both quantitative and qualitative image quality in myocardial delayed enhancement CT compared to conventional HIR. By reducing image noise and enhancing the visibility of myocardial scars, DLIR may improve the diagnostic confidence. These findings support the incorporation of DLIR into routine MDE-CT protocols and highlight its potential to expand the clinical utility of cardiac CT for myocardial tissue characterization. Future studies are warranted to validate these results in larger, multi-center cohorts and to explore the impact of DLIR on diagnostic accuracy and clinical outcomes.