In this prospective study we included eight consecutive children and young adults with overall ten knees and known or suspected synovitis of the knee joint who underwent routine diagnostic MRI between October 2016 and July 2019 at our institution. The patients were recruited at the department of Pediatric and Adolescent Medicine, consulting the Special Ambulance for Pediatric Rheumatology. Inclusion criteria were age between 0 till 25 years, consulting the Special Ambulance with suspected or having arthritis of the knee joint or follow up, indication for contrast-enhanced MRI and informed consent by them or their parents. Exclusion criteria were contraindication for 3 Tesla MRI – like implanted medical devices such as pace makers, neural stimulaters or insulin pumps, shell splinters, contraindication for application of constrast agent and missing informed consent. For comparison, we meanwhile enrolled 5 healthy young adults by posting in the University and University Hospital of Ulm who underwent contrast-free MR imaging of both knees. Inclusion criteria were age between 18 till 30 years, not having any known inflammatory pathologies of the knee joints and informed consent. Exclusion criteria were contraindication for 3 Tesla MRI – like implanted medical devices such as pace makers, neural stimulaters or insulin pumps, shell splinters and missing informed consent.

We conducted all study work in accordance with the Helsinki Declaration. The prospective study was approved by the institutional ethics committee (Nr. 320/16). Also, we obtained informed written consent from all participants or legal guardians for diagnostic procedures.

All patients underwent clinical routine MRI on the same 3.0 Tesla scanner MAGNETOM Skyra (Siemens Healthcare, Erlangen, Germany). Standard sequences were scanned as needed for clinical diagnostic workup (e.g. T1-weighted turbo spin echo transverse, proton-density-weighted with fat saturated sagittal). Diffusion-weighted scans were acquired prior to injection of contrast agent using the following parameters: transverse segmented-readout multi-shot DWI (RESOLVE, Siemens Healthcare); repetition time (TR) = 3580 ms, echo time (TE) 1 = 57 ms, TE 2 = 88 ms; 5 readout segments; b-values 0/50/100/150/ 200/300/400/600/800/1000 s/mm2; fat suppression; trace-weighted; GeneRalized Autocalibrating Partial Acquisition (GRAPPA) with integrated Parallel Acquistion Techniques (iPAT) = 2; 3D Diagonal diffusion mode; monopolar gradient scheme; one average at b = 0, 50, 100, 150, 200, 300 s/mm2, two averages at b = 400 und 600 s/mm2, three averages at b = 800 und 1000 s/mm2; echo-planar imaging factor 63; voxel size 1.3 × 1.3 × 3.0 mm3; field of view 180 mm; acquisition time 5 min 56 s. Apparent diffusion coefficient (ADC)0/1000 maps were automatically calculated by mono-exponential curve-fitting on the MRI console. After intravenous injection of one weight-adapted standard dose of Gadolinium-based contrast agent (children under 18 years of age mostly goderate meglumine = Dotarem, Guerbet, Paris, France; over 18 years of age and one girl aged 11 years Gadovist, Bayer, Leverkusen, Germany), we acquired post-contrast transverse T1-weighted turbo spin echo with fat saturation, followed by coronal or sagittal T1-weighted scans when clinically indicated. Eight children/young adults and the five participants of the control group were examined in supine position feet-first and with a dedicated 15-channel transmit/receive knee coil of the manufacturer. In two girls (2 and 17 years old), both knees were examined in supine position head-first with an 18-channel body coil of the manufacturer in place, the first one following sedation administered by a paediatric anaesthesiologist.

The IVIM model describes the total diffusion-weighted signal as a superposition of signal decay [16]. The perfusion-related signal contributes a fraction f to the unweighted (b = 0 s/mm2) signal S0 and decays exponentially with a so-called pseudo-diffusion coefficient Dp. In contrast, signal undergoing true diffusion is subject to exponential decay with the true diffusion coefficient D. IVIM describes the total diffusion-weighted signal intensity S as a bi-exponential function:

$$\textrm=}_0\left(\left(1-\textrm\right)\ }^}+\textrm\ }^}\right)$$

Usually, the pseudo-diffusion coefficient Dp is approximately one order of magnitude larger than the tissue diffusion coefficient D. This means that the perfusion-related signal is more attenuated as compared to the diffusive signal for a given b value. At high b values (commonly ≥200 s/mm2) the perfusion-related signal is very low and may be neglected. In this case, the diffusion-weighted signal is described by:

$$}_}=}_0\left(1-\textrm\right)}^}$$

Fitting of the parameters D and f was done on a standard personal computer with custom written code in the software MATLAB® (version R2015b; MathWorks®, Natick, MA). Areas with signal intensity lower than 30% of maximum intensity on the unweighted (b = 0 s/mm2) images were masked out on the D and f parameter maps.

Parameter fitting was performed with the commonly used “segmented” two-step fit: first, D and f were calculated from images with b-values above a certain threshold b-value (here called b-threshold), where the influence of the pseudo-diffusion was negligible (assuming D* = 0), simplifying the fitting model to a mono-exponential signal decay function. Subsequently, these D- and f-values were used for the perfusion coefficient calculation in the bi-exponential model.

Qualitative image analysis was performed on a dedicated radiological workstation with commercially available PACS software (IMPAX EE R20.Ink, AGFA HealthCare, Mortsel, Belgium). A board-certified pediatric radiologist with 12 years of experience in pediatric extra-cranial DWI - also trained in this special DWI reading in knee joint synovium - (first reader) and a senior radiologist trained in pediatric imaging for 4 years (second reader) - but with no specific training in reading this DWI technique in joint synovium - reflecting different levels of experience, independently completed all qualitative image analysis for the patient group. Each reader was blinded to all clinical patient data and to the results of the other reader (see also Fig. 1). All contrast enhanced images and then three weeks later all diffusion-weighted images at b = 1000 s/mm2 with the corresponding ADC map in random order were assessed separately. Presence and degree of synovitis were recorded by each reader on a modified 5-item Likert scale, which is shown in Table 1. The categories 3 and 4 were considered to represent manifest synovitis/arthritis, while category 1 was thought to exclude the presence of manifest synovitis, based on imaging criteria. Category 2 indicated mild synovial irritation. The respective subjective level of diagnostic confidence (LoC) was documented along with each rating on a modified 3-item Likert-scale (1 = undecided, questionable, low confidence; 2 = intermediate, diagnosis established with some confidence; 3 = certain diagnosis, high confidence). The transverse contrast-enhanced T1w (ce-T1w) scan was considered the diagnostic standard of comparison.

Flowchart of study patient knees examined independently by the first and second reader for qualitative image analysis. The scoring of the contrast-enhanced (= ce) and diffusion-wheighted (= dwi) images could be finally analyzed for all 10 patient knees

Blinded to patient data and to the results of preceding qualitative image analysis, a resident radiologist with two years of experience in pediatric radiology performed quantitative analysis of patient and proband data using polygonal regions of interest (ROIs) manually and separately drawn on representative cross-section of diffusion-weighted images and the corresponding ADC map, which were placed side-by-side on the monitor. Quantitative data measured for synovium, joint effusion (if present) and muscle included signal intensity as the mean of two measurements on DWI b = 1000 s/mm2. Using custom-written code in the software MATLAB, the IVIM parameters perfusion fraction (f) and diffusion coefficient (D) were calculated from the total segmented tissue volume as mean value and standard deviation, minimum and maximum values. Quantitative analysis was repeated by the same reader after several weeks for assessment of intra-observer variability.

Normally distributed data are presented as mean ± standard deviation. We used the Mann-Whitney test to compare means of two independent data sets deviating from normal distribution. Kappa statistics were calculated to assess inter-observer agreement. All data analysis were performed with SPSS Version 27 for Windows (IBM, Armonk, NY).