Suha Faleh Alharbi, Fatma Aboul-Enein, Hassan B. Hawsawi, Sara Alharbi, Ahmed Samman, Khulood Jamal Abusafieh, Rami Alharbi, Technical Evaluation of Automated Left Ventricular Quantification in Cardiac Magnetic Resonance Imaging: Comparison with Manual Reference Measurements, Journal of Big Data Research, Volume 1, Issue 3, 2026, Pages 32-43, ISSN 2768-0207, https://doi.org/10.14302/issn.2768-0207.jbr-26-6477.
(https://openaccesspub.org/big-data-research/article/technical-evaluation-of-automated-left-ventricular-quantification-in-cardiac-magnetic-resonance-imaging-comparison-with-manual-reference-mea-2404)
Abstract: Background Automated left ventricular (LV) quantification may reduce the manual post-processing burden of cardiac magnetic resonance (CMR) imaging but requires evaluation against manual reference measurements before routine clinical use. Purpose To evaluate automated CMR post-processing for LV quantification against manual reference measurements and to compare its performance with Inline scanner-derived measurements. Methods CMR examinations from 178 patients at a single tertiary referral center were retrospectively evaluated. Ejection fraction (EF), end-diastolic volume (EDV), end-systolic volume (ESV), stroke volume (SV) and myocardial mass obtained with automated syngo.via post-processing and Inline scanner analysis were compared with manual expert measurements using intraclass correlation coefficients (ICC), Bland–Altman analysis, root mean square error (RMSE) and mean absolute percentage error (MAPE). Results Automated post-processing showed excellent agreement with manual measurements for all five parameters (ICC 0.957–1.000; MAPE 0.72–3.21%). Bland–Altman mean biases were small: −0.83 percentage points for EF, 0.74 mL for EDV, −1.43 mL for ESV, −1.45 mL for SV and −1.03 g for myocardial mass, although ESV showed wider limits of agreement. Inline measurements showed greater disagreement with the manual reference, including an EF bias of +4.22 percentage points and moderate ICCs for SV (0.733), EDV (0.673) and ESV (0.619), with consistently higher RMSE and MAPE. Conclusions Automated CMR post-processing agreed closely with manual reference measurements and showed lower measurement error than Inline scanner-derived measurements, supporting its use to aid routine LV quantification with expert review where segmentation is uncertain. Further multicenter evaluation across scanners and patient populations is warranted.
Keywords: cardiac magnetic resonance; left ventricular quantification; automated segmentation; ejection fraction; ventricular volumes; Bland–Altman analysis; intraclass correlation coefficient