Reproducibility of Perfusion-Weighted and Diffusion-Weighted Magnetic Resonance Imaging in Patients with Ocular Tumours Authors: Pieter A.M. Janssen, Teresa A. Ferreira, T.H. Khanh Vu, Jan-Willem M. Beenakker, Lisa Klaassen Published: 2026 ABSTRACT AbstractIntroductionDiffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI) can assess biological tumour characteristics of uveal melanoma (UM), the most common primary intraocular malignant tumour in adults. Recent studies propose that these techniques can aid in the differential diagnosis, provide prognostic information, and enable early assessment of treatment response following radiation therapy. However, it is important to determine the extent to which any observed differences can be attributable to true biological changes in the tumour rather than day-to-day physiological or measurement variations. Therefore, this study aimed to assess the reproducibility of ocular DWI and PWI measurements between two centres.MethodIn this prospective study, 15 patients with UM were included. Each patient underwent two MRIs at two different centres with a mean interval of 28 days (range: 12–43 days). Apparent diffusion coefficient (ADC), relative peak intensity, and outflow percentage of the tumour were compared between the two timepoints. Mean absolute differences between the two timepoints were calculated.ResultsMean absolute differences between the two timepoints were 0.07 × 10−3 mm2/s (95% CI, 0.04–0.11) for ADC, 0.11 (95% CI, 0.04–0.18) for relative peak intensity, and 6 (95% CI, 3–10) percentage points for outflow percentage.ConclusionThis study shows that DWI and PWI parameters demonstrate reproducibility in UM patients between two different centres, with variabilities smaller than differences typically observed in differential diagnosis or treatment response assessments. Introduction Uveal melanoma (UM) is the most common primary intraocular malignancy in adults, with an incidence of approximately 6 cases per one million individuals [1]. Conventional imaging modalities for diagnosis and follow-up include ultrasound, fundoscopy, and fluorescein angiography [2]. Treatment options range from enucleation to eye- and vision-preserving therapies, such as episcleral brachytherapy and proton beam therapy (PBT) [1]. In the last decade, the use of magnetic resonance imaging (MRI) in ocular oncology has increased, enabling three-dimensional tumour visualization [3–7] and noninvasive assessment of tumour biology with functional imaging techniques such as diffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI) [2, 8, 9]. With DWI, the tissue’s apparent diffusion coefficient (ADC) can be assessed, which is a quantitative biomarker of water mobility based on the tissue microenvironment. PWI uses an intravenous contrast agent to assess tumour perfusion, with analysis focussing on the type of time-intensity curve (TIC), as well as TIC-derived measurements such as peak intensity and outflow percentage [10, 11]. Different clinical applications of these functional imaging techniques have been proposed, often complementing the conventional ophthalmic imaging modalities. For example, PWI and DWI allow further characterization of UM, which may aid in differential diagnosis [8, 10, 12]. In PWI, UM often shows a wash-out TIC, whereas progressive TICs, which are more prevalent in benign lesions, have not been reported [8]. Furthermore, Tang et al. [11] highlighted the role of PWI in response assessment, with favourable PWI changes observed after radiotherapy, before changes in size were apparent. In a study by Foti et al. [13], a significant increase in ADC was shown after radiotherapy. Additionally, quantitative analysis of PWI has been proposed to potentially differentiate between high- and low-risk genetic sub-classes of UM, potentially providing a biopsy-less prognostication [8, 14–17]. However, before these quantitative metrics can be used to further guide clinical decision-making, it is essential to determine whether the observed differences reflect true biological changes or merely normal day-to-day variability or inter-scanner differences [18]. Currently, the reproducibility of functional MRI for ocular tumours remains unknown. Therefore, in this study, we aimed to assess the reproducibility of perfusion-weighted and diffusion-weighted MRI in patients with UM. Methods Study Design In this study, parameters from PWI and DWI sequences were compared for UM patients who were eligible for PBT. This analysis was part of a larger, prospective study, in which patients received several MRI scans in the context of scientific research, in addition to the two scans that are performed in clinical care at our centre. Patients Between October 2022 and February 2024, 15 patients diagnosed with UM and eligible for PBT were included in this study at centre 1, based on the inclusion and exclusion criteria and willingness to participate in the study. The diagnosis of UM was made by an ocular oncologist, based on fundoscopy, fundus photography, ultrasound, and fluorescein angiography. Eligibility for PBT was based on national guidelines including larger tumours defined as a tumour prominence ≥7 mm and basal diameter ≥16 mm, or for juxtapupillary located tumours. All patients were referred to centre 2 for PBT, where they received four fractions of 15 GyE on four consecutive days [19]. Exclusion criteria included contraindications for MRI, such as claustrophobia, MRI-incompatible implants, or pacemakers. For the analysis in this study, patients were also excluded if the tumour prominence was <3 mm, as the ADC measurement can be unreliable due to partial volume effects [10]. Each patient underwent two MRI scans, including DWI and PWI sequences. These scans were performed over a period of 2 to 6 weeks. The first MRI was conducted before clip placement surgery at centre 1 and was part of regular clinical care. The second MRI was performed at the first appointment at centre 2, at least 7 days after clip placement surgery and before PBT. MRI Protocol All MRIs were acquired using a dedicated 3-tesla ocular MRI protocol with a 4.7-cm surface receive coil. At centre 1, a 3 T Ingenia Elition MRI was used, while at centre 2 scans were performed with a 3 T Ingenia MRI (both software version 5.6, Philips Healthcare, Best, the Netherlands). The entire scan protocol contained anatomical scans, a dynamic contrast-enhanced scan and the diffusion-weighted scan, and was previously described in more detail by Ferreira et al. [10]. Scans were optimized to limit the effect of potential eye movement [14]. Characteristics of the DWI and PWI scans analysed in this study are summarized in Table 1. The protocols at centre 1 and centre 2 differed slightly due to differences in gradient performance and availability of specific software options (Table 1). The angulation of the DWI scans at the second timepoint was chosen such that it matched those of the first timepoint. For PWI, a bolus of 0.1 mmol/kg bodyweight of gadoterate meglumine (centre 1: Clariscan®, GE Healthcare, IL, USA and centre 2: Dotarem®, Guerbet Diagnostic Imaging, Villepinte, France) was administered using a power injector with an injection rate of 2 mL/s. Table 1.PWI and DWI scan parametersScan nameVoxel size (mm3)FOVEcho train length/FA (deg)TE (ms)/TR (ms)Scan time (mm:ss)Fat. suprAdditional parametersDWI (centre 1)0.63 × 0.63 × 2.4100 × 100 × 2288/9050/1,55501:33SPIRB = 0, 800 s/mm2, spin echoDWI (centre 2)0.63 × 0.63 × 2.4100 × 100 × 2288/9047/1,80201:48SPIRB = 0, 800 s/mm2, spin echoPWI (centre 1)1 × 1 × 1.6280 × 80 × 321/132.3/4.503:59Proset2 s/dynamic, TWIST scan accelerationPWI (centre 2)1 × 1 × 1.5280 × 80 × 321/132.3/4.504:55Proset3.8 s/dynamicFOV, field of view; TE, echo time; TR, repetition time; SPIR, spectral presaturation with inversion recovery; TWIST, time-resolved angiography with stochastic trajectory sequence [20]. PWI and DWI scan parameters FOV, field of view; TE, echo time; TR, repetition time; SPIR, spectral presaturation with inversion recovery; TWIST, time-resolved angiography with stochastic trajectory sequence [20]. Image Analysis The DWI analysis was performed using Sectra IDS7 (Sectra AB, Linköping, Sweden, version 25.2), with the ADC map as provided by the scanner software. The PWI analysis was performed with Vitrea (Vital Images, Minnetonka, USA, version 7.16.0.703). For both DWI and PWI, a single-slice circular region of interest (ROI) was manually drawn within the tumour by one observer (P.J.). These ROIs were reviewed, and if necessary adjusted, by an ocular MRI expert with over 12 years of experience (J.W.B.). This was done on the tumour slice with the largest lesion. The same ROI diameter size was applied for both DWI and PWI in each patient. ROI placement was guided by corresponding T1- and T2-weighted imaging to differentiate the tumour from retinal detachment, necrosis, and inflammation [10, 14, 19, 21]. For tumours with multiple components, the most malignant measurement was chosen (e.g., lowest ADC value for DWI, and highest outflow and/or enhancement for PWI). The same size ROI was placed on the same location across both MRIs for each patient. For PWI, the tumour was semi-quantitatively assessed by relative peak signal intensity and outflow percentage (Fig. 1). The relative peak signal intensity was calculated by dividing the peak signal by the baseline signal averaged over the first 20 s (Fig. 1). Outflow percentage was defined as the relative difference between peak intensity and the intensity 2 min post-peak (Fig. 1). At all timepoints, ROI placement was verified to be fully in the tumour and adjusted if necessary to account for eye motion. Determination of DWI and PWI parameters. a ADC map with ROI (red circle). b One of the timepoints of the dynamic scan with ROI (red circle). c TIC with relative peak intensity and outflow percentage. Statistical Analysis The primary aim was to assess functional MRI reproducibility by comparing ADC values, relative peak intensity, and outflow percentages between both timepoints per patient. Mean absolute differences were calculated, and Bland-Altman plots were used for comparison. Furthermore, the reproducibility coefficient (RDC) was calculated for the ADC and relative peak intensity, by using the within-subject coefficient of variation (wCV), following the recommendations of the Quantitative Imaging Biomarkers Alliance (QIBA) [18]. Results Fifteen patients were included in this study, all of whom underwent two MRI scans. One patient was excluded for both the PWI and DWI analyses due to a haemorrhage within the tumour between both timepoints. Furthermore, 2 patients did not complete the PWI scan at the second timepoint due to patient preferences or difficulty gaining access to the blood vessel and were therefore excluded for the PWI analysis. Consequently, a total of 14 patients were included in the DWI analysis, and 12 patients in the PWI analysis, with a mean interval between both scans of 28 ± 8.8 (range: 12–43) days. Patient and tumour characteristics are summarized in Table 2. Table 2.Baseline characteristics of the subjectsCharacteristicsPatients (n = 14)Age, years57±14 (range: 25–72)Sex – male57%Eye – OD50%AJCC classification: T1/T2/T3/T4a0 (0%)/3 (21%)/10 (71%)/1 (7%)Shape: mushroom/dome3 (21%)/11 (79%)Prominence at baseline, mm7.1±2.6 (range: 3.5–12)Change in prominence between timepoints, mm0.4±0.8 (range −0.5 to 3.0)Data are presented as percentage (%) for categorical variables and as mean ± SD for continuous variables. aAccording to the 8th American Joint Committee on Cancer (AJCC) edition [22]. Baseline characteristics of the subjects Data are presented as percentage (%) for categorical variables and as mean ± SD for continuous variables. aAccording to the 8th American Joint Committee on Cancer (AJCC) edition [22]. Mean ADC values were 1.00 ± 0.12 × 10−3 mm2/s for the first timepoint and 1.03 ± 0.08 × 10−3 mm2/s for the second timepoint, with a mean absolute difference between both timepoints of 0.07 ± 0.06 × 10−3 mm2/s (95% CI: 0.04–0.11 mm2/s) (Fig. 2). The wCV was 0.066 mm2/s, corresponding to an RDC of 18.4%. Bland-Altman plots for ADC, relative peak intensity, and outflow percentage. To determine the difference, the values at timepoint 1 were subtracted from the values at timepoint 2. Mean relative peak intensity values were 1.92 and 2.00, for the first and second timepoints, respectively, with a mean absolute difference of 0.11 ± 0.10 (95% CI: 0.04–0.18) (Fig. 2). The wCV was 0.046, corresponding to an RDC of 12.7%. For the outflow percentage, mean values were −18% at the first timepoint and −17% at the second timepoint, with a mean absolute difference of 6 ± 6 (95% CI, 3–10) percentage points (Fig. 2). Examples of 2 patients at both timepoints are shown in Figure 3. Individual patient data are provided in online supplementary Table 1 (for all online suppl. material, see https://doi.org/10.1159/000552801). ADC map and TIC for a representative patient (a) and a patient with large differences in outflow percentage (b). Discussion This study evaluated the reproducibility of DWI and PWI for patients with UM. The ADC values observed in this study align with previous findings for UM [4, 10, 13, 15, 21]. The PWI peak enhancement of our cohort is slightly higher than those reported by Ferreira et al. and Buerk et al. [10, 23], likely due to the studied patient population, which included patients treated with PBT, which are typically patients with a larger or juxtapapillary tumour not suitable for brachytherapy. In the context of differential diagnosis, the reproducibility observed of 0.07 × 10−3 mm2/s for ADC, 0.11 for relative peak intensity, and 6 percentage points for outflow percentage are substantially smaller than the differences reported between UM and benign lesions. For instance, Buerk et al. [23] observed a 0.4 lower peak intensity in benign choroidal lesions, and the review by Jaarsma et al. [8] highlighted larger differences in ADC values between benign and malignant entities, for example, 1.11 × 10−3 mm2/s for UM, compared to values of around 1.4 × 10−3 mm2/s for inflammatory lesions and around 1.6 × 10−3 mm2/s for other benign orbital lesions. This supports the clinical relevance of functional MRI metrics and their potential utility in distinguishing tumour types. Similarly, studies evaluating treatment response with DWI or PWI reported larger differences after treatment than the observed reproducibility, except for smaller differences in outflow percentage observed early after PBT. For example, Foti et al. [13] noted substantial changes in ADC values after PBT, for example, from 1.03 to 1.26 × 10−3 mm2/s at 1 month post-treatment, a change threefold larger than the reproducibility. Furthermore, Tang et al. [11] described an average absolute decrease of more than 0.5 in relative peak intensity, which is five times larger than the method’s reproducibility. The same study used a threshold of 5% for a visually distinguishable difference in outflow percentage after treatment, which corresponds to the reproducibility found in this study. Therefore, differences larger than 5% can most likely be attributed to true biological changes and not just day-to-day physiological or measurement variations. All in all, this shows that the reproducibility observed in this paper is sufficient for differential diagnosis and treatment response assessment. In this study, we focus on reproducibility under the premise of realistic clinical patient re-evaluation in a multi-centre setting, i.e., patients returning for a follow-up scan. Hence, possible differences in patient, equipment, and eye position are expected. A reliable MRI biomarker needs to be robust to such deviations. Therefore, a reproducibility analysis was chosen instead of a repeatability analysis. There are multiple factors that could affect the reproducibility. An important one is the consistency in the identification of the ROI between the different MRI scans. UMs can be heterogeneous [10], resulting in different values based on the placement of the ROI. In this study, the ROIs were drawn manually by one person, as in clinical practice. Even though all ROIs used in this study were checked by an expert with 10+ years of experience in ocular MRI, we can expect that the ROI placements will not be perfectly consistent between timepoints. However, a perfect match of manually defined ROIs also cannot be expected in clinical practice. Therefore, it is useful to incorporate potential inconsistencies in region definition into account in assessing these scans in clinical practice. In this context, 3D ROI determination could improve reproducibility, but might not be an accurate representation of the current clinical workflow [24]. Furthermore, the two MRI scans compared were performed at two different centres with MR scanners from the same vendor and with comparable scan protocols. On average, both the ADC and peak intensity were slightly higher at the second centre. Differences between the scan protocols included different temporal resolutions for the PWI scans and the use of time-resolved angiography with stochastic trajectory sequence scan acceleration in centre 1, which was not available at centre 2. Both these factors may have an effect on, for example, peak intensity. Therefore, reproducibility within the exact same scanner might be higher than the reproducibility in this study. However, the reproducibility between different centres with scanners from different vendors might be lower. Limitations One of the main limitations of the study is the possible effect of the clip placement surgery, which was performed between both timepoints. Additionally, the relatively long interval between scans, on average 28 days, may have allowed for biological changes. Although UM is considered a slow growing tumour [20, 22], some patients did show an increase in tumour prominence. These factors could lead to an overestimation of the variability and thus affect the assessment of reproducibility. Nonetheless, the observed reproducibility remains more than sufficient for current clinical applications, such as differential diagnosis and treatment response assessment. Another limitation of this study is the small sample size, which reduced statistical power and prevented subgroup analyses, such as assessing reproducibility across tumour sizes. Furthermore, the cohort consisted exclusively of patients referred for PBT. Future studies with larger and more diverse populations could validate these findings and examine reproducibility across different clinical subgroups. Conclusion This study shows that DWI and PWI parameters demonstrate good reproducibility in UM patients between two different centres, with variability smaller than differences typically observed in differential diagnosis or treatment response assessment. These findings support the use of functional MRI parameters as imaging biomarkers for these purposes in both clinical practice and research. Statement of Ethics This prospective study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Local Ethics Committee (registration number NL7343305820). Written informed consent was obtained from all participants. Conflict of Interest Statement The C.J. Gorter MRI Center (LUMC), to which J.-W.M. B. and L.K. are affiliated, receives research support from Philips Healthcare. Funding Sources This research was funded by Varian, a Siemens Healthineers company (HollandPTC-Varian Consortium grant ID [project 2020016]). Author Contributions Conception, design, data acquisition, analysis, and interpretation: P.A.M.J., L.K., and J.-W.M.B. Drafting or reviewing for important intellectual content and final approval of the final version to be published: T.H.K.V. and T.A.F.