Quality assessment of cardiovascular magnetic resonance in the setting of the European CMR registry: description and validation of standardized criteria

Abstract

Background: Cardiovascular magnetic resonance (CMR) has become an important diagnostic imaging modality in cardiovascular medicine. However, insufficient image quality may compromise its diagnostic accuracy. We aimed to describe and validate standardized criteria to evaluate a) cine steady-state free precession (SSFP), b) late gadolinium enhancement (LGE), and c) stress first-pass perfusion images. These criteria will serve for quality assessment in the setting of the Euro-CMR registry.

Methods: Thirty-five qualitative criteria were defined (scores 0-3) with lower scores indicating better image quality. In addition, quantitative parameters were measured yielding 2 additional quality criteria, i.e. signal-to-noise ratio (SNR) of non-infarcted myocardium (as a measure of correct signal nulling of healthy myocardium) for LGE and % signal increase during contrast medium first-pass for perfusion images. These qualitative and quantitative criteria were assessed in a total of 90 patients (60 patients scanned at our own institution at 1.5T (n=30) and 3T (n=30) and in 30 patients randomly chosen from the Euro-CMR registry examined at 1.5T). Analyses were performed by 2 SCMR level-3 experts, 1 trained study nurse, and 1 trained medical student.

Results: The global quality score was 6.7±4.6 (n=90, mean of 4 observers, maximum possible score 64), range 6.4-6.9 (p=0.76 between observers). It ranged from 4.0-4.3 for 1.5T (p=0.96 between observers), from 5.9-6.9 for 3T (p=0.33 between observers), and from 8.6-10.3 for the Euro-CMR cases (p=0.40 between observers). The inter- (n=4) and intra-observer (n=2) agreement for the global quality score, i.e. the percentage of assignments to the same quality tertile ranged from 80% to 88% and from 90% to 98%, respectively. The agreement for the quantitative assessment for LGE images (scores 0-2 for SNR <2, 2-5, >5, respectively) ranged from 78-84% for the entire population, and 70-93% at 1.5T, 64-88% at 3T, and 72-90% for the Euro-CMR cases. The agreement for perfusion images (scores 0-2 for %SI increase >200%, 100%-200%,<100%, respectively) ranged from 81-91% for the entire population, and 76-100% at 1.5T, 67-96% at 3T, and 62-90% for the Euro-CMR registry cases. The intra-class correlation coefficient for the global quality score was 0.83.

Conclusions: The described criteria for the assessment of CMR image quality are robust with a good inter- and intra-observer agreement. Further research is needed to define the impact of image quality on the diagnostic and prognostic yield of CMR studies.

Figures

Figure 1

Quality Evaluation of CMR Images.Total qualitative score: sum of qualitative scoring for SSFP images (12 criteria: range of scores 0–19, for LGE images (10 criteria: range of scores 0–19), and for perfusion images (13 criteria: range of scores 0–20). Total range: 0–60. Quantitative LGE score: 5 parameters measured yielding scores of 0–2 (SNR <2; 2–5; >5, respectively) for each the anterior and inferior LV wall. Total range of mean scores: 0–2. Quantitative perfusion score: 5 parameters measured yielding scores of 0–2 (%SI increase >200%; 100-200%; <100%, respectively) for each the anterior and inferior LV wall. Total range of mean scores: 0–2. Total quantitative score: sum of quantitative LGE and perfusion score. Range: 0–4. Global quality score: sum of total qualitative and total quantitative score: Range: 0–64.

Figure 2

Wrap around in a cine SSFP sequence: Chest wall (located outside the field of view) is projecting onto the left ventricle (red arrows).

Figure 3

Image blurring/mis-triggering in cine a SSFP sequence: Respiratory motions, mis-triggering of the R-wave or irregular heartbeats induce a blurred aspect of the image (red arrows).

Figure 4

Artifact in a cine SSFP sequence caused by ferromagnetic material: Sternotomy wires locally disturb the magnetic field (red arrows). However, it is not considered a significant artifact in this case, since it does not extend onto the LV.

Figure 5

Shimming artifact in a cine SSFP sequence: Magnetic field inhomogeneities produce dark band and flow related (red arrows) artifacts on the LV.

Figure 6

Respiratory ghost (indicated by red lines on the right image, B) in a LGE sequence: Respiratory motion during the image acquisition projects replicates of the chest wall onto the LV.

Figure 7

Cardiac ghost (indicated by red lines on the right image, B) in a LGE sequence: Cardiac motion during the acquisition is seen in this case as multiple replicates of LV contours in the phase-encoding direction.

Figure 8

Wrap around artifact in a LGE sequence: A structure outside the field of view is projected onto the LV (red arrows).

Figure 9

ECG mis-triggering in a LGE sequence: Image quality is decreased by both, a mis-triggering artifact and by cardiac ghosts. According the definitions of artifact criteria, the most severe artifact, i.e. the mis-triggering artifact, is considered for scoring only.

Figure 10

Distribution of qualities in the entire study population (n=90) for the global quality score (A) as well as for the qualitative (B) and quantitative scores (C, D). The first tertile of quality scores (score <9) encompasses the largest portion of studies ranging between 69% - 78% of all cases (for readers 1–4). A similar distribution is observed for the qualitative and the quantitative perfusion score, whereas the LGE score shows a considerable portion of increased signal in the normal myocardium of 21-24% (SNR 2–5) and 28-32% (SNR >5) of all studies indicating a sub-optimal myocardial signal nulling in these cases.

Figure 11

Inter-observer agreement for the global quality score. It ranges from 83% to 98% for the 4 readers (n=90 studies). As the level of quality deteriorates (ascending black line indicating mean global quality score of all 4 readers, units to the right of the figure), the level of agreement slightly declines, while for the good quality examinations at 1.5T (1.5T group to the left with a mean score of 4.0), the agreement between the 4 readers is excellent ranging from 97% to 100%.

Figure 12

The various levels of agreement between experts (Exp), a trained study nurse (Nurse), and a trained medical student (Stud) is shown. For all comparisons, moderate to substantial agreements were found.