As a multi-site virtual cancer institute, Cancer Core Europe requires standardized imaging-biomarkers for therapy monitoring in multicenter trials. Usually therapy monitoring is based on CT images and RECIST evaluation, although multiparametric functional MRI offers several advantages for the non-invasive characterization of tumor viability. This approach is rarely adopted due to a lack of objective standardization. In a recently submitted study, we suggest that a standardized MRI protocol including diffusion weighted MRI (DWI) is feasible and we also quantified the residual variability of measurement parameters. This was done by phantom and volunteer measurements (single-shot T2w and DW-EPI) at the seven CCE sites using the slightly heterogeneous MR-hardware produced by three different vendors. Repeated measurements were performed at the sites and across them, comparing qualitative and quantitative results including a Radiomics – analysis, thus constituting an emerging medical category of image post processing methods that extract multiple features from radiographic medical images using data-characterization algorithms.
Our results are very encouraging: for DWI/ADC phantom measurements, the maximum deviation between centers decreased by 2 %. In volunteers, the measurement variation in 2 repeated scans did not exceed 11% for ADC and is below 20% for single-shot T2w in systematic liver ROIs. The measurement variation between sites amounted to 20% for ADC and < 25% for single-shot T2w and is therefore in keep with variations noted for interrater RECIST evaluations. Concerning radiomics, the classification experiments and visual separation worked well, though better for ADC than for the ultra-fast T2w. Thus, the harmonization of MR acquisition and post processing parameters between our centers resulted in acceptable standard deviations for MR/DWI-imaging. Therefore, multiparametric MRI can be considered a valuable tool in oncoming trials and overcoming the limitations of RECIST–based response evaluation.
One of the goals of CCE’s Imaging Task Force is to establish quantitative imaging-biomarkers through radiomics features extracted by computational methods. The first step we identified is the comparability of images extracted from the different scanners through phantom studies. Due to the fact that current research points to the important role for radiomics in both diagnosis and response assessment, reliable quantitative features are needed, and even small inter-scanner variations and poor repeatability may impede the use of radiomics in multicenter studies.
We circulated a CT dedicated phantom with homogeneous and heterogeneous regions of interest across the different centers, and analyzed images obtained through a variety of acquisition and reconstruction parameters using the same methodology. Our results showed that three radiomic features exhibit a low inter-scanner variability and would allow multicenter studies: GLCM_entropy, GLCM_dissimilarity, and GLZLM_ZLNU. Moreover, the discriminating capacities to distinguish tissue heterogeneities was also obtained through GLCM dissimilarity. Finally, we observed a high variability in the computation of radiomic features in case different filters were applied during the acquisition and reconstruction process.
We concluded that radiomics multicenter studies seem feasible by fixing the acquisition and reconstruction parameters and applying a restricted number of feature parameters only. This study also warrants a closer collaboration with CT manufacturers in the standardization of radiomics research, ideally through the development of a consensual CT acquisition protocol devoted to radiomics, with specific reconstruction filters common to all CT manufacturers.