Pediatric whole-body magnetic resonance imaging: comparison of STIR and T2 Dixon sequences in the detection and grading of high signal bone marrow changes

Whole-body MRI is increasingly being used to diagnose and monitor diseases, such as chronic non-bacterial osteomyelitis, malignant/metastatic disease, Langerhans cell histiocytosis, syndromes, and genetic predisposition with increased tumor risk, and more recently, juvenile idiopathic arthritis [1]. However, standardized and validated imaging protocols, as well as reference standards by age for the growing skeleton, are lacking. A recent systematic review showed that the typical whole-body MRI protocol in children includes fluid-sensitive sequences only, or fluid sensitive along with T1-weighted (T1W) sequences, but the technical settings varied considerably across institutions [2]. Of fluid sensitive sequences, all being fast spin echo or turbo spin echo sequences, a short time inversion recovery (STIR) was by far the most common, being well known for its robust fat suppression. Alternative techniques including T2-weighted (T2W) Dixon were used/mentioned in only two out of the 54 studies [2]. In contrast, the Dixon technique is well-established in adult radiology, with a variety of musculoskeletal applications [3].

Several techniques enable fat signal suppression, based on (1) difference in resonance frequency with water by means of frequency selective pulses: fat saturation (fat-sat) techniques; (2) phase contrast techniques, (3) short T1 relaxation time by means of inversion recovery sequences (STIR technique), (4) Dixon method, and (5) hybrid techniques combining several of these fat suppression techniques such as spectral pre-saturation with inversion recovery [4]. The different fat suppression techniques have their advantages and disadvantages; while STIR is less sensitive to B0- and B1-heterogeneity and provides additional T1 and T2 contrast, it has a lower signal to noise due to the inversion delay. Using the Dixon technique, on the other hand, provides a better signal-to-noise ratio (SNR), and generates “water-only,” out of phase, in phase, and “fat-only” images simultaneously, and provides additional information on anatomy and water/fat content. The sequence is hampered with a specific artifact, the Dixon swap; however, recent software updates have reduced this problem.

In adults, the T2W Dixon and STIR techniques have been compared in a few studies showing diverging results [5, 6], while no such studies have been published for children. We have previously shown that grading high signal bone marrow intensity on a 0–2 scale performs well based on T2W Dixon images [7]. The purpose of this study was to compare T2W Dixon and STIR in the detection and grading of high signal intensity bone marrow changes at 1.5 T, on whole-body MRI in healthy children.

A total of 77 whole-body MRI examinations in 77 healthy children (36 males) between 6 and 19 years (mean age 12 years (SD 3.1)) were included (Fig. 3). A total of 28,100 images were analyzed. All children had at least one area of high signal (median 6 areas, range 1–30). A total of 127/3311 (3.8%) anatomical areas were excluded due to artifacts or suboptimal SNR, of which 78 (61.4%) were based on STIR alone, 23 (18.1%) were based on both STIR and Dixon images, and 16 (12.6%) were based on Dixon alone. Except for two, the excluded areas were shoulder, elbow, hand, or foot.

A total of 568 lesions were identified and scored during the initial readings, of which 6 (1.1%) in the spine, 8 (1.4%) in the thoracic cage, 58 (10.2%) in the upper extremities, 41 (7.2%) in the pelvis, and 435 (76.6%) in the lower extremities. On the 3-reader consensus with all sequences and Dixon reconstructions available (substitute gold standard), 23/568 (4.1%) lesions (2 in the upper extremities and 21 in the pelvis/lower extremities) were judged insignificant, thus should not have been scored, leaving 545 “true” lesions for further analysis.

Based on the substitute gold standard, 490 of the 545 (89.9%) “true” lesions were visible on both sequences, while 27 (5.0%) were visible on STIR only and 28 (5.1%) were visible on T2W Dixon only (Table 2) (Fig. 4, Fig. 5). The rates of “true” lesions visible on both sequences ranged from 75% for the pelvis to 92.9% for the upper extremities (Table 2) while corresponding percentages for the axial skeleton and thoracic cage could not be estimated due to low numbers (p < 0.001).

The proportion of true positive high signal lesions was higher for the T2W Dixon images as compared to STIR (74.2% (95% CI = 70.5–77.8%) vs. 68.2% (95% CI = 64.3–72.2%) (p = 0.029), while the proportion of false negatives was lower (25.9% (95% CI = 22.2–29.5%) vs. 31.7% (95% CI = 31.7–27.8%)) (p = 0.035) for T2W Dixon and STIR, respectively) (Fig. 6).

The initial agreement between T2W Dixon and STIR for grading of signal intensity, i.e., based on the 545 “true” lesions was poor, with a kappa value of − 0.02, increasing to 0.34 (95% CI = 0.25–0.42) after dichotomizing into a 0–1 score (Table 3).

Extension of the high signal areas was scored for 185 of the 545 identified “true” high intensity signal areas. There was a moderate agreement between the T2W TSE Dixon and STIR-based scores on a 0–4 scale, with a kappa value of 0.45 (95% CI = 0.34–0.56) (Table 4). Excluding lesions with an extension below 5% did not improve the agreement (kappa value 0.45 (95% CI = 0.33–0.57)).

We have shown, in a large cohort of healthy children and adolescents, that 90% of high signal bone marrow areas identified on a 1.5-T whole-body MRI were seen on both STIR and water-only T2W Dixon, while 5% were seen on STIR only and 5% were seen on T2W Dixon only. Nearly all signal changes seen exclusively on one sequence had low conspicuity/signal intensity. T2W Dixon-based readings identified a 6% higher number of high signal areas, a higher proportion of true positives, and fewer false negatives than did STIR. Grading signal intensity on a 0–2 scale showed poor agreement; although this to some extent is due to inherent subjective reader variability, we believe it reflects that the two sequences display high signal areas differently.

The fact that nearly all areas of medium to fluid signal were visible on both scans when re-scored in consensus by three experienced radiologists (substitute gold standard) implies that high signal areas of clinical importance most likely will be identified on both scan types. Our results indicate that in follow-ups of known pathology in a particular patient, as well as when searching for multiplicity, the same protocol should be used. The higher sensitivity and specificity in identifying high signal areas, as well as the additional information provided by Dixon, favor this sequence over STIR, as does the fact that most of the images excluded for further analysis based on artifacts or suboptimal signal-to-noise ratio were STIR-weighted. The latter should be interpreted with caution, however, as the STIR sequence was performed at the end of the examination, with an increased risk of movement artifacts in a bored child.

Although the majority of the 545 “true” lesions were identified on both sequences, conspicuity, or signal intensity, was perceived and scored differently, with T2W Dixon identifying a slightly higher number of high signal areas. This might reflect differences between the two sequences beyond voxel size, image artifacts, and reading environments. We believe that the inconsistent initial grading as displayed in the contingency table (Table 3) is caused by the slightly lower SNR in STIR images (Fig. 2), small movements between scans (Fig. 4, Fig. 5), the subjective nature of the scoring, the large number of localizations assessed per whole-body MRI, or to a combination of these issues. Moreover, the choice of parameters might have played a role. Ideally, a comparison between the two sequences should have included a repeatability exercise for each of the two sequences, separately, to help identify, and thus exclude disagreement due to precision issues. However, based on 96 examinations from the same WB MRI study, we have previously shown that scoring high intensity bone marrow changes on a 0–2 scale performs well, both for the same, and between readers based on T2W Dixon images [7]. We assume that the precision of STIR-based readings does not differ substantially from the Dixon-based readings, an assumption supported by others [10]. Moreover, the technical settings used in our study were harmonized between the sequences. We therefore argue that a substantial part of the inconsistency seen between STIR and T2 Dixon-based readings of signal intensity is explained by true differences between the sequences.

In adults, several studies have compared the diagnostic performance of T2W Dixon and STIR sequences for focused examinations. In their study of 22 patients, mean age 80.9 years, Heynen et al. compared T2W Dixon, T1W, and STIR images in the assessment of radiographically occult fractures to the femoral neck, using a 3-T scanner [10]. Interestingly, the interobserver agreement was significantly lower for the T2W Dixon images than for the STIR images, with kappa values of 0.70–0.79 and 0.87–0.93, respectively. Moreover, using information for all available sequences as a substitute gold standard, T2W Dixon water-only images had lower sensitivity and accuracy for identification of fractures than had STIR [10]. Their results contrast those of others, for example in the assessment of the lumbar spine [11], the hand [5], and the sacro-iliac joints [12]. In sum, the published results vary significantly, reflecting different scanners and technical settings, as well as different study designs. Similar studies in children, or studies addressing whole-body MRI, do, to the best of our knowledge, not exist.

We found a moderate agreement between Dixon and STIR images for the assessment of signal extension, when based on 185 of the 545 high intensity areas. Again, the inconsistencies may be due to issues as listed for signal intensity, such as differences in SNR, small movements between scans, and the subjective nature of the scoring.

Again, we would argue that our results have implications for clinical practice, underscoring the importance of applying identical whole-body MRI protocols during follow-up of children with metastatic or inflammatory bone marrow lesions, particularly due to the high frequency of silent lesions [13, 14]. In our study, 5% of 545 lesions were visible on STIR only, and another 5% were visible on T2 Dixon only, based on a consensus score between three experienced readers. Similar differences most likely occur between MRI scanners from different vendors, sequences with different scan parameters, and between scanners of different field strengths. Compared to the 1.5-T scanner used in our study, a 3-T scanner has higher signal-to-noise ratio, wider chemical shift between the fat- and water-signal peak, higher specific absorption rate, larger susceptibility effects with resultant artifacts, greater B0 and B1 heterogeneity, and a longer T1 relaxation time. These differences affect the reliability and consistency of fat suppression, and, in theory after optimization of the technical settings, should provide better images, thus better observer agreement. In a busy environment, issues like these might be ignored due to availability constraints on MRI laboratories, among others.

We acknowledge several limitations to our study. First, our results apply to whole-body 1.5-T MRIs in healthy children only. However, a high proportion of the high signal intensity areas were grade 2, e.g., with a moderate to fluid like signal intensity, which in a clinical setting might mimic pathology [15, 16]. Second, subjective grading of signal intensity on fat-suppressed T2W images is hampered with difficulties; however, we have recently shown that grading intensity of high signal areas within the bone marrow performs well after meticulous calibration [7]. Third, blinding according to sequence was impossible and fourth, our choice of sequence parameters is only one of many options, and the conclusion holds for only these particular sequence parameters. The strengths include the meticulous calibration between observers prior to scoring, the relatively high numbers and balanced distribution of findings, and similar technical settings for both scans.

We have shown, in a large cohort of healthy children and adolescents, that most high signal bone marrow changes identified on a 1.5-T whole-body MRI were seen on both STIR and water-only T2W Dixon, and that nearly all signal changes seen exclusively on one sequence had low conspicuity/signal intensity. T2W Dixon-based readings identified a 6% higher number of lesions, a higher proportion of true positives, and fewer false negatives than did STIR. This underscores the importance of using identical protocols when following bone-marrow signal changes over time. The additional information provided by the Dixon-w images, as well as the better SNR, however, favors the Dixon sequence.