Whole-body magnetic resonance imaging (MRI) is increasingly being used for a number of indications. Our aim was to review and describe indications and scan protocols for diagnostic value of whole-body MRI for multifocal disease in children and adolescents, we conducted a systematic search in Medline, Embase and Cochrane for all published papers until November 2018. Relevant subject headings and free text words were used for the following concepts: 1) whole-body, 2) magnetic resonance imaging and 3) child and/or adolescent. Included were papers in English with a relevant study design that reported on the use and/ or findings from whole-body MRI examinations in children and adolescents. This review includes 54 of 1,609 papers identified from literature searches. Chronic nonbacterial osteomyelitis, lymphoma and metastasis were the most frequent indications for performing a whole-body MRI. The typical protocol included a coronal STIR (short tau inversion recovery) sequence with or without a coronal T1-weighted sequence. Numerous studies lacked sufficient data for calculating images resolution and only a few studies reported the acquired voxel volume, making it impossible for others to reproduce the protocol/images. Only a minority of the included papers assessed reliability tests and none of the studies documented whether the use of whole-body MRI affected mortality and/or morbidity. Our systematic review confirms significant variability of technique and the lack of proven validity of MRI findings. The information could potentially be used to boost attempts towards standardization of technique, reporting and guidelines development.

Background Chronic nonbacterial osteomyelitis (CNO) is a rare autoinflammatory bone disorder. Little information exists on the use of imaging techniques in CNO. Materials and methods We retrospectively reviewed clinical and MRI findings in children diagnosed with CNO between 2012 and 2018. Criteria for CNO included unifocal or multifocal inflammatory bone lesions, symptom duration >6 weeks and exclusion of infections and malignancy. All children had an MRI (1.5 tesla) performed at the time of diagnosis; 68 of these examinations were whole-body MRIs including coronal short tau inversion recovery sequences, with additional sequences in equivocal cases. Results We included 75 children (26 boys, or 34.7%), with mean age 10.5 years (range 0–17 years) at diagnosis. Median time from disease onset to diagnosis was 4 months (range 1.5–72.0 months). Fifty-nine of the 75 (78.7%) children presented with pain, with or without swelling or fever, and 17 (22.7%) presented with back pain alone. Inflammatory markers were raised in 46/75 (61.3%) children. Fifty-four of 75 (72%) had a bone biopsy. Whole-body MRI revealed a median number of 6 involved sites (range 1–27). Five children (6.7%) had unifocal disease. The most commonly affected bones were femur in 46 (61.3%) children, tibia in 48 (64.0%), pelvis in 29 (38.7%) and spine in 20 (26.7%). Except for involvement of the fibula and spine, no statistically significant differences were seen according to gender. Conclusion Nearly one-fourth of the children presented with isolated back pain, particularly girls. The most common sites of disease were the femur, tibia and pelvic bones. Increased inflammatory markers seem to predict the number of MRI sites involved.

Background Whole-body magnetic resonance imaging (MRI) is increasingly being used in children, however, to date there are no studies addressing the reliability of the findings. Objective To examine intra- and interobserver reliability of a scoring system for assessment of high signal areas within the bone marrow, as visualized on T2-weighted, fat-saturated images. Materials and methods Ninety-six whole-body MRIs (1.5 T) in 78 healthy volunteers (mean age: 11.5 years) and 18 children with chronic nonbacterial osteomyelitis (mean age: 12.4 years) were included. Coronal water-only Dixon T2-weighted images were used to score the left lower extremity/pelvis for high signal intensity areas, intensity (0–2 scale), extension (0–4 scale) and shape and contour in a blinded fashion by two pairs of radiologists. Results For the pelvis, grading of bone marrow signal showed moderate to good intra- and interobserver agreement with kappa values of 0.51–0.94 and 0.41–0.87, respectively. Corresponding figures for the femur were 0.61–0.68 within and 0.32–0.61 between observers, and for the tibia 0.60–0.72 and 0.51–0.73. Agreement for assessing extension was moderate to good both within and between observers for the pelvis (k = 0.52–0.85 and 0.35–0.80), for the femur (0.52–0.67 and 0.51–0.60) and for the tibia (k = 0.59–0.69 and 0.47–0.63) except for the femur metaphysis/diaphysis, with interobserver kappa values of 0.29–0.30. Scoring of shape was moderate to good within observers, but in general poorer between observers, with kappa values of 0.40–0.73 and 0.18–0.69, respectively. For contour, the corresponding figures were 0.35–0.62 and 0.09–0.54, respectively. Conclusion MRI grading of intensity and extension of high signal intensity areas within the bone marrow of pelvis and lower limb performs well and thus can be used interchangeably by different observers, while assessment of shape and contour is reliable for the same observer but is less reliable between observers. This should be considered when performing clinical trials.

Background Manual assessment of bone marrow signal is time-consuming and requires meticulous standardisation to secure adequate precision of findings. Objective We examined the feasibility of using deep learning for automated segmentation of bone marrow signal in children and adolescents. Materials and methods We selected knee images from 95 whole-body MRI examinations of healthy individuals and of children with chronic non-bacterial osteomyelitis, ages 6–18 years, in a longitudinal prospective multi-centre study cohort. Bone marrow signal on T2-weighted Dixon water-only images was divided into three color-coded intensity-levels: 1 = slightly increased; 2 = mildly increased; 3 = moderately to highly increased, up to fluid-like signal. We trained a convolutional neural network on 85 examinations to perform bone marrow segmentation. Four readers manually segmented a test set of 10 examinations and calculated ground truth using simultaneous truth and performance level estimation (STAPLE). We evaluated model and rater performance through Dice similarity coefficient and in consensus. Results Consensus score of model performance showed acceptable results for all but one examination. Model performance and reader agreement had highest scores for level-1 signal (median Dice 0.68) and lowest scores for level-3 signal (median Dice 0.40), particularly in examinations where this signal was sparse. Conclusion It is feasible to develop a deep-learning-based model for automated segmentation of bone marrow signal in children and adolescents. Our model performed poorest for the highest signal intensity in examinations where this signal was sparse. Further improvement requires training on larger and more balanced datasets and validation against ground truth, which should be established by radiologists from several institutions in consensus.