Improvement of Quantification of Myocardial Synthetic ECV with Second-Generation Deep Learning Reconstruction

Area of Science:

• Cardiovascular Imaging
• Medical Physics
• Radiology

Background:

• Synthetic extracellular volume (ECV) quantification offers an alternative to methods requiring hematocrit values.
• High-quality cardiac CT images are crucial for accurate synthetic ECV measurements.
• Second-generation Deep Learning Reconstruction (DLR) enhances cardiac CT image quality, reducing noise and increasing resolution.

Purpose of the Study:

• To compare the accuracy of synthetic ECV quantification using four different CT image reconstruction methods.
• To evaluate hybrid iterative reconstruction (HIR), model-based iterative reconstruction (MBIR), DLR, and second-generation DLR for synthetic ECV measurement.
• To determine which reconstruction method provides the most accurate synthetic ECV quantification compared to laboratory-derived ECV.

Main Methods:

• Retrospective analysis of 80 patients undergoing cardiac CT scans.
• Derivation cohort (n=40) established a linear regression model between blood test hematocrit and right atrial CT values.
• Validation cohort (n=40) calculated synthetic hematocrit and ECV, then assessed correlation and agreement with laboratory ECV across four reconstruction methods.

Main Results:

• All four reconstruction methods demonstrated a strong correlation between synthetic ECV and laboratory ECV (R ≥ 0.95, p < 0.001).
• Second-generation DLR exhibited the lowest bias (-0.20) and limit of agreement (2.35) in Bland-Altman analysis.
• HIR, MBIR, and DLR showed higher bias and limits of agreement compared to second-generation DLR.

Conclusions:

• Second-generation DLR provides the most accurate synthetic ECV quantification among the evaluated methods.
• This advanced reconstruction technique minimizes bias and limits of agreement, enhancing clinical utility.
• Second-generation DLR is recommended for precise synthetic ECV measurements in cardiac CT.