Image reconstruction impacts haemodynamic parameters derived from 4D flow magnetic resonance imaging with compressed sensing

Abstract

Aims: 4D blood flow measurements by cardiac magnetic resonance imaging (CMR) can be used to simplify blood flow assessment. Compressed sensing (CS) can provide better flow measurements than conventional parallel imaging (PI), but clinical validation is needed. This study aimed to validate stroke volume (SV) measurements by 4D-CS in healthy volunteers and patients while also investigating the influence of the CS image reconstruction parameter λ on haemodynamic parameters.

Methods and results: Healthy participants (n = 9; 20-62 years) underwent CMR with 2D, 4D-CS, and 4D-PI flow. Patients (n = 30, 17 with congenital heart defect; 2-75 years) had 4D-CS added to their clinical examination. Impact of λ was assessed by reconstructing 4D-CS data for six different λ values. In healthy volunteers, 4D-CS and 4D-PI SV differed by 0.4 ± 6.5 mL [0.6 ± 9.1%; intraclass correlation coefficient (ICC) 0.98], and 4D-CS and 2D flow by 0.9 ± 7.0 mL (0.9 ± 10.6%; ICC 0.98). In patients, 4D-CS and 2D flow differed by -1.3 ± 6.0 mL (-7.2 ± 20%; ICC 0.97). SV was not dependent on λ in patients (P = 0.75) but an increase in λ by 0.001 led to increased differences between 4D-CS and 4D-PI of -0.4% (P = 0.0021) in healthy participants. There were significant differences for ventricular kinetic energy (systole: P < 0.0001; diastole: P < 0.0001) and haemodynamic forces (systole: P < 0.0001; diastole: P < 0.0001), where error increased with increasing λ values in both healthy participants and patients.

Conclusion: 4D flow CMR with CS can be used clinically to assess SV in paediatric and adult patients. Ventricular kinetic energy and haemodynamic forces are however sensitive to the change in reconstruction parameter λ, and it is therefore important to validate advanced blood flow measurements before comparing data between scanners and centres.

Keywords: cardiac magnetic resonance imaging (CMR); congenital heart disease; haemodynamic forces; kinetic energy; paediatric.

Graphical Abstract

Figure 1

SV in healthy participants. Bland–Altman plots show differences between SV assessed by (A) 2D flow and 4D-CS, (B) 2D flow and 4D-PI volumes, and (C) 4D-CS and 4D-PI. Both 4D-CS and 4D-PI agree with 2D flow to a similar extent as repeated 2D or 4D flow acquisitions (cf. Results section). (D) Bland–Altman plot showing difference in SV between 2D flow and 4D flow with CS in 29 patients, and a total of 97 vessels.

Figure 2

Impact of change in reconstruction parameter on SV in the ascending aorta. (A) Comparison of 4D-CS and 4D-PI and different reconstruction parameters in eight healthy volunteers and (B) 4D-CS with default setting (λ = 0.0015) and different reconstruction parameters in six patients. Values are presented as median and interquartile range. The agreement between 4D-CS and 4D-PI was not affected by change in reconstruction parameter.

Figure 3

Impact of change in reconstruction parameter on ventricular kinetic energy and haemodynamic forces in healthy volunteers (n = 6). Difference in peak ventricular kinetic energy (A and B) and haemodynamic forces (C and D) in the left ventricle during systole (A and C) and diastole (B and D) between 4D-CS and 4D-PI with different reconstruction parameters. Values are presented as median and interquartile range. Both ventricular kinetic energy and haemodynamic forces were lower in 4D-CS than in 4D-PI with default reconstruction parameter (λ = 0.0015). The reconstruction parameter λ has a significant impact on the parameters.

Figure 4

Visualization of ventricular kinetic energy (KE) in the left ventricle during early diastole in a patient. Images show peak KE with (A) λ = 0.0005, (B) λ = 0.0015, and (C) λ = 0.0030. Peak KE decreased with as λ increased. (D) KE across the cardiac cycle in the same patient with different reconstruction parameters. The asterisk and vertical dashed line show the time frame of peak diastolic KE where the images were generated.

Figure 5

Impact of change in reconstruction parameter on ventricular kinetic energy and haemodynamic forces in patients (n = 6). (A) and (B) show the difference in peak ventricular kinetic energy, and (C) and (D) show differences in haemodynamic forces in the left ventricle between 4D-CS with default setting (reconstruction parameter λ = 0.0015) and 4D-CS with different reconstruction parameters, in six patients. (A) and (C) show systolic values, and (C) and (D) show diastolic values. Values are presented as median and interquartile range. The reconstruction parameter λ has a significant impact on the values for both ventricular kinetic energy and haemodynamic forces.

Figure 6

Impact of change in reconstruction parameter on ventricular kinetic energy and haemodynamic forces in healthy participants (n = 8). Difference in peak ventricular kinetic energy (A and B) and haemodynamic forces (C and D) in the left ventricle during systole (A and C) and diastole (B and D) between 4D-CS and default setting (reconstruction parameter λ = 0.0015) and 4D-CS with different reconstruction parameters. Values are presented as median and interquartile range. The reconstruction parameter λ has a significant impact on the values for both ventricular kinetic energy and haemodynamic forces.