Rapid Fire Session

Session: Rapid Fire Session 1: Peds/CHD Flow and Vascular Disease (Monitor 10)

The challenge of slow flow: Need for improvement of the accuracy of pulmonary venous flow determination by 2D, 4D and 5D flow cardiac magnetic resonance

Thursday, February 5, 2026

10:22 AM - 10:29 AM

Location: Galapagos / Aruba / Ilhabela (2nd Lower Level)

Presenting Author(s)

Fabienne Dirbach, MD

Physician, PhD student
Lausanne University Hospital and University of Lausanne
Lausanne, Vaud, Switzerland

Primary Author(s)

Fabienne Dirbach, MD

Physician, PhD student
Lausanne University Hospital and University of Lausanne
Lausanne, Vaud, Switzerland

Co-Author(s)

ET

Estelle Tenisch, MD

Physician
University Hospital (CHUV) and University of Lausanne (UNIL), Switzerland
MP

Milan Prsa, MD

Cardiologist
University Hospital (CHUV) and University of Lausanne (UNIL), Switzerland
SF

Sara Faessler

MS
Lausanne University Hospital and University of Lausanne, Vaud, Switzerland
Christopher W. Roy, PhD

Lecturer
Lausanne University Hospital (CHUV) and University of Lausanne (UNIL)
Lausanne, Vaud, Switzerland
Mariana Falcão, PhD

PhD
University Hospital (CHUV) and University of Lausanne (UNIL)
Lausanne, Vaud, Switzerland
PG

Paolo Garelli, MSc

PhD student
University Hospital (CHUV) and University of Lausanne (UNIL), Switzerland
EA

Efena Akporeha, MSc

PhD Candidate
Lausanne University Hospital
Lausanne, Vaud, Switzerland
DR

David Rodrigues, MSc

Deputy chief radiology technician
Lausanne University Hospital and University of Lausanne, Switzerland
MM

Michael Markl, PhD

Professor
Northwestern University Feinberg School of Medicine
Chicago, Illinois, United States
MS

Matthias Stuber, PhD

Professor/Director
CIBM/CHUV/UNIL
Lausanne, Switzerland
Davide Piccini, PhD

Collaboration Manager
Siemens Healthineers
Lausanne, Vaud, Italy
Juerg Schwitter, MD

Full Professor
University Hospital and University of Lausanne
Lausanne, Vaud, Switzerland
Tobias Rutz, MD

Medical doctor
University Hospital and University of Lausanne
Lausanne, Vaud, Switzerland

Background: Accurate pulmonary venous (PV) flow quantification by cardiovascular magnetic resonance (CMR) is essential in patients with complex congenital heart disease (CHD). Standard 2D phase-contrast (2D flow) remains the clinical reference but is limited by challenges in image plane prescription and measurement difficulties in small vessels¹. 4D and 5D flow sequences allow comprehensive volumetric and time-resolved acquisition but may suffer from segmentation difficulties and low signal in small veins. Iron-based contrast agents (ferumoxytol, FE) are discussed to improve image signal. This study compares PV flow quantification using standard 2D flow, an accelerated respiratory-navigated 4D flow sequence², and a free-running, fully self-gated 5D flow technique.^3,4

Methods: In CHD patients 2D, 4D and 5D flow data were acquired. Velocity encoding (VENC, cm/s) was adapted for 2D flow to 20% higher than expected maximum velocity and for 4D and 5D flow to 200 cm/s. Flow was determined in the main, right and left pulmonary artery (MPA, RPA, LPA), left and right pulmonary veins (RPV, LPV). 3D vessel segmentation was performed for 4D and 5D flow datasets. PV net flow volumes (NFV) were compared between the three techniques and to respective arterial NFV for evaluation of internal consistency. Influence of non-contrast, gadolinium-enhanced (GD), and FE acquisitions was determined.

Results: 59 CHD patients (mean age 28±13y, 42% women, aortopathy 10, right-sided CHD 30, cyanotic CHD, 2, Fontan 4, Ross 10, shunt lesion 3) were included. PV flow was acquired by 2D flow for 21 and in all patients for 4D and 5D flow.

4D and 5D flow underestimated NFV in pulmonary arteries and particularly in PV compared to 2D flow. The bias between 2D and 4D flow NFV was significantly smaller than between 2D and 5D (Table 1).

Internal consistency comparison of all three techniques demonstrated large biases and wide confidence intervals (Figure). 4D flow yielded slightly narrower CI, while 5D flow showed particularly large CI when evaluating internal consistency of right-sided PV. The presence or not of contrast agents did not influence the differences of NFV between 2D, 4D and 5D flow (Table 2).

Conclusion: All three techniques lack accuracy for PV flow determination in a typical CHD cohort. 4D and 5D flow underestimated PV flow volumes while 4D flow demonstrated slightly better agreement with 2D flow and better internal consistency than 5D flow. Advanced sequences hold promise for improved flow assessment, but our findings underscore that current techniques remain limited, and iron-based contrast did not significantly enhance accuracy. The choice of the VENC could play a role and the implementation of a dual VENC technique could further improve the accuracy of flow measurements by 4D and 5D flow techniques. Other factors like lack of post-processing software with dedicated segmentation algorithms for PV, probably also influence PV flow quantification.

Table 1 Comparison of bias (ml/beat) 2D vs. 4Dflow to bias 2D vs 5D flow (all patients)

Net volume (ml/beat)	2D vs 4D	2D vs 5D	4D vs 5D	p
MPA	-8.62 (-41.24, 24.01) *	-6.9 (-39.41, 25.61) *	1.87 (-26, 29.74) *	0.003
RPA	-10.25 (-26.1, 5.59)^†	-8.64 (-25.18, 7.91)^†	1.87 (-10.86, 14.59)^†	< 0.001
LPA	-8.08 (-28.47, 12.32)^#	-8.02 (-30.07, 14.03) ^#	0.19 (-15.48, 15.86)^#	< 0.001
RPV	-6.44 (-20.63, 7.75)^‡	-17.69 (-38.97, 3.58)^‡	-8.91 (-25.85, 8.02)^‡	< 0.001
LPV	-9.47 (-36.11, 17.18)	-10.70 (-32.06, 10.66)^§	-3.61 (-21.07, 13.84)^§	0.035

Data shown are bias (ml/beat) (95% lower, upper limit of agreement). P-values refer to analyses of variance (ANOVA) with Tukey post-hoc analyses for comparison of bias between the three sequences.

*p=0.17 2Dvs4D vs 4Dvs5D, p=0.002 2Dvs5D vs 4Dvs5D

^†p < 0.001 2Dvs4D vs 4Dvs5D and 2Dvs5D vs 4Dvs5D

^#p < 0.001 2Dvs4D vs 4Dvs5D, p< 0.001 2Dvs5D vs 4Dvs5D

^‡p < 0.001 2Dvs4D vs 2Dvs5D, p=0.004 2Dvs5D vs 4Dvs5D

^§p=0.023 2Dvs5D vs 4Dvs5D

Abbreviations: MPA = main pulmonary artery, RPA = right pulmonary artery, LPA = left pulmonary artery, RPV= right pulmonary vein, LPV= left pulmonary veinTable 2: Comparison of differences of pulmonary venous net flow volumes between 2D, 4D and 5D flow as expressed by the biases depending on contrast mediums (all patients)

Net volume (ml/beat)	No contrast	Gadolinium	Ferumoxytol	p
RPV 2D vs 4D	-13.79	-7.4 (-11.28, -3.52)	0.5 (-16.34, 17.34)	0.131
RPV 2D vs 5D	-25.84 (-138.86, 87.19)	-18.31 (-24.2, -12.42)	-9.39 (-37.99, 19.20)	0.242
RPV 4D vs 5D	-11.12 (-15.36, -6.89)	-8.08 (-11.74, -4.41)	-7.14 (-12.15, -2.12)	0.432
LPV 2D vs 4D	-12.73	-10.79 (-19.14, -2.45)	-2.19 (-28.99, 24.62)	0.620
LPV 2D vs 5D	-16.31 (-28.06, -4.55)	-11.01 (-18.1, -3.93)	-5.5 (-15.74, 4.74)	0.569
LPV 4D vs 5D	-3.93 (-7.62, -0.23)	-3.54 (-7.36, 0.28)	-3.18 (-8.57, 2.21)	0.981

Data shown are bias (ml/beat) (95% lower, upper limit of agreement). P-values refer analyses of variance (ANOVA) of comparison of bias between the three modalities of contrast/no contrast.

Abbreviations: see table 1Figure: Bland-Altman analyses for comparison of internal consistency for bias of net volume (ml/beat): p-values refer to ANOVA with Tukey post-hoc analyses for comparison of bias between the three sequences. RPV vs RPA: p<0.001 4D vs 5D Abbreviations: see table 1 Patients with Fontan-type physiology were excluded to avoid the potential influence of collaterals.