Automated image quality assessment for selecting among multiple magnetic resonance image acquisitions in the German National Cohort study

In magnetic resonance imaging (MRI), the perception of substandard image quality may prompt repetition of the respective image acquisition protocol. Subsequently selecting the preferred high-quality image data from a series of acquisitions can be challenging. An automated workflow may facilitate and improve this selection. We therefore aimed to investigate the applicability of an automated image quality assessment for the prediction of the subjectively preferred image acquisition. Our analysis included data from 11,347 participants with whole-body MRI examinations performed as part of the ongoing prospective multi-center German National Cohort (NAKO) study. Trained radiologic technologists repeated any of the twelve examination protocols due to induced setup errors and/or subjectively unsatisfactory image quality and chose a preferred acquisition from the resultant series. Up to 11 quantitative image quality parameters were automatically derived from all acquisitions. Regularized regression and standard estimates of diagnostic accuracy were calculated. Controlling for setup variations in 2342 series of two or more acquisitions, technologists preferred the repetition over the initial acquisition in 1116 of 1396 series in which the initial setup was retained (79.9%, range across protocols: 73–100%). Image quality parameters then commonly showed statistically significant differences between chosen and discarded acquisitions. In regularized regression across all protocols, ‘structured noise maximum’ was the strongest predictor for the technologists’ choice, followed by ‘N/2 ghosting average’. Combinations of the automatically derived parameters provided an area under the ROC curve between 0.51 and 0.74 for the prediction of the technologists’ choice. It is concluded that automated image quality assessment can, despite considerable performance differences between protocols and anatomical regions, contribute substantially to identifying the subjective preference in a series of MRI acquisitions and thus provide effective decision support to readers.


SUPPLEMENTAL MATERIAL
Table S1.Criteria catalog for visual image quality ratings in the NAKO MRI study (translated from version 02 of the internal documentation, example images omitted).Scores were assigned according to a 3-point Likert scale: 1) 'excellent' image quality not impaired by artifacts, images appropriate for data post-processing [marked below as Green]; 2) 'good' image quality with limited impairment by artifacts, images appropriate for data post-processing [marked below as Yellow]; 3) 'poor' image quality due to artifacts or insufficient coverage, images generally not appropriate for post-processing [marked below as Red].The protocols used for functional or quantitative imaging (Resting State EPI BOLD, MOLLI SAX, and Multiecho 3D VIBE) were not rated.

Neurodegenerative Focus T1w 3D MPRAGE
Minimal coverage Left-right: from ear to ear; dorso-ventral: entire brain (high parietal region: all layers up to the skull included) to the lower border of the cerebellum; rostro-caudal: entire brain from frontal to occipital pole.

Figure S1 .FigureFigure S2 .Figure
Figure S1.ROC curves from regularized regression of the combined set of image quality parameters with the outcome 'chosen vs. discarded acquisition': a across all protocols on 1,000 bootstrap samples, b across all protocols on 1,000 bootstrap samples (excluding the parameter 'specific SNR' to minimize missing data), c-m for individual protocols (three protocols had an insufficient sample size for inclusion: Resting State EPI BOLD, PDw FS 3D SPACE, and T2w 2D FSE).AUC with 95% CI corresponds to mean AUC and respective percentiles from the distribution over all bootstrap samples.Left to right: LASSO regression, Elastic Net regression, ridge regression.

Table S1 .
(continued)Green if completely covered (Optimal coverage: Complete imaging from the shoulder girdle above the clavicle to the middle of the femur bone.)Minimum differentiable structures Liver, portal vein, spleen, splenic vein, pancreas, adrenal glands, kidneys, renal pelvis, renal vein, visceral fat Other Considerations SWAP artifacts (swap of fat/water voxels).Yellow if in visceral non-target organs (e.g., bladder) or 1-20% of visceral/subcutaneous fat of the torso • Green if, for example, in the area of the extremities B0 inhomogeneities (if caused by ECG cables/electrodes, still green; otherwise, proceed as for SWAP artifacts)?Breathing artifacts?Foldover artifacts?
• Red if only partially covered or not covered at all • Yellow if narrowly covered • Green if completely covered • Red if completely wrong orientation of planes (e.g., missing a cardiac chamber) or completely not captured • Yellow if still consistent with the correct orientation of long-axis sections (e.g., LVOT in the 4CV slightly cut) • Green if correctly captured Minimum differentiable structures Myocardium can be differentiated Other Considerations Pulsation artifacts or magnetic field inhomogeneity or banding artifacts: Red if target structures (left and right ventricles) are affected or Yellow if they are still distinguishable.Yellow is also given if the atria are not evaluable Cine SSFP SAX Minimal coverage Heart: Base to apex depicted; at least 1 complete cardiac cycle • Red if only partially covered or not covered at all • Yellow if narrowly covered • Green if completely covered • Red if not completely covered • Yellow if narrowly covered or if the lung apex is covered in the last slice.• • Red if in visceral target organs (lung, mediastinum, liver, pancreas, kidneys) or >20% of visceral/subcutaneous fat of the torso •

Table S1 .
(continued)Green in the absence of artifacts but also if, for example, only subcutaneous fat tissue appears incompletely saturated • Red if no fat saturation is present or if fat saturation did not work in the target organs (e.g., around the sacroiliac joint) • Yellow if fat saturation in non-target organs is inadequate (e.g., signal-rich bone marrow in the ischial bones) • • Red if not completely covered in terms of the number of vertebrae • Yellow if vertebrae are not completely captured laterally or if neuroforamina are not fully captured in scoliosis •

Table S1 .
(continued)Minimum differentiable structures Longitudinal ligaments, interspinal ligaments, intervertebral disc spaces, neuroforamina T1-12 and their contents, spinal cord and conus, facet joint space or cartilage, dorsal muscle fascia Red if not completely covered in terms of the number of vertebrae • Yellow if vertebrae are not completely covered laterally or if neuroforamina are not fully captured in scoliosis • Green if completely covered Minimum differentiable structures Longitudinal ligaments, interspinal ligaments, intervertebral disc spaces, neuroforamina L1-S1 and their contents, in addition to nerve roots S2-4, conus and caudal fibers, facet joint space or cartilage, dorsal muscle fascia, height of the aortic bifurcation, pre-sacral fat tissue Other Considerations Red if ventral saturator overlaps vertebral structures, Yellow if dorsal subcutaneous fat is depicted inhomogeneously •