ESTRO 37 Abstract book

S63

ESTRO 37

biology of tumours may differ between (metastatic) lesions inside a patient and finally also at the intra- tumour (‘voxel’) level. Multi-parametric imaging could play a role for tumour classification for predicting treatment outcome, several examples will be shown.

to qMRI of 30 patients prior to prostatectomy. Using generalized linear mixed-effect modelling, we built a predictive model for spatial mapping of high-grade GP, incorporating the inter- and intra-patient heterogeneity. Our results show the potential of qMRI for localizing high- grade GP on voxel-level. We will further discuss that the final pathology of patients can accurately be derived from the mapping of high-grade GP inside the prostate lesions. We will also demonstrate the accuracies achieved for the prediction of high-grade GP on voxel-level as well as high-grade prostate cancer, on patient-level. We previously discussed the potential for dose differentiations in the prostate gland considering Gleason Pattern based-heterogeneity in the radiosensitivity (1). If by spatial mapping of high-grade GP inside the lesions we improve the accuracy of targeted biopsies, we can further use this information for dose differentiations purposes. 1. Ghobadi G, de Jong J, Hollmann BG, van Triest B, van der Poel HG, Vens C, et al. Histopathology-derived modeling of prostate cancer tumor control probability: Implications for the dose to the tumor and the gland. Radiother Oncol 2016;119:97–103. SP-0129 Multi-parametric imaging for outcome prediction and response assessment E. Malinen 1 1 Oslo University Hospital / University of Oslo, Department of Medical Physics, Oslo, Norway Abstract text Non-invasive medical imaging holds great promise for improving non-invasive assessment of solid cancers before and during radiotherapy and subsequent prediction of response to treatment. Imaging methods that look beyond anatomical features and provide information on tumor biology, and also organ function, are of particular interest. Here, imaging modalities such as positron emission tomography (PET) or magnetic resonance imaging (MRI) provide image-based parameters that are significant associated with treatment outcome in single- parameter (univariate) analysis. Examples are standardized uptake value (SUV) and metabolic tumor volume (MTV) from PET and volume transfer rate (K trans ) and apparent diffusion coefficient (ADC) from MRI. In this respect, hybrid PET/MR scanning may be used to extract many of these parameters in one imaging session. Although links have been established between these single image parameters and response/outcome, prediction accuracy is seldom high and there is an urgent need to further develop such image-based markers. In addition to having access to large-scale image data, preferably acquired at many institutions, there could be several approaches to advance image-based prediction models. Of these are 1) model-driven and 2) data-driven methods. In model-driven approaches, an hypothesis is formed based on the underlying biological meaning of image data, and the resulting ‘mechanistic’ model is tested against outcome data. One example of such a model could be combining blood flow and cell density- related parameters from MRI, assumed related to hypoxia and tumor burden, respectively. Data-driven methods in the context of imaging and radiotherapy outcome are called Radiomics, and will not be further discussed. The talk will present published and ongoing work on multi- parametric imaging in the context of outcome prediction and response assessment, highlight promising applications and address limitations and challenges.

Figure 1: The processing supervoxel workflow: For each patient, the multi-parametric images are normalized and summed, and used to create supervoxels (top row). For each supervoxel, the median of all image features is calculated. Next, the supervoxels of all patients are combined and clustered based on their image features. Patients are assigned to the different clusters and differences in overall survival between the groups can be assessed. SP-0128 Multi-parametric functional imaging for lesion identification and RT personalisation in prostate cancer G. Ghobadi 1 , P.J. Van Houdt 1 , J. De Jong 2 , I. Walreven 1 , C.V. Dinh 1 , H.G. Van der Poel 3 , S.W. Heijmink 4 , F.J. Pos 1 , S. Isebaert 5 , R. Oyen 6 , S. Raylander 7 , L. Bentzen 8 , S. Høyer 9 , E.M.E. Klawer 1 , C. Dinis Fernandes 1 , K. Tanderup 8 , K. Haustermans 5 , U. Van der Heide 1 1 Netherlands Cancer Institute, Radiotherapy Department, Amsterdam, The Netherlands 2 Netherlands Cancer Institute, Pathology Department, Amsterdam, The Netherlands 3 Netherlands Cancer Institute, Urology Department, Amsterdam, The Netherlands 4 Netherlands Cancer Institute, Radiology Department, Amsterdam, The Netherlands 5 University Hospitals Leuven, Radiation Oncology Department, Leuven, Belgium 6 University Hospitals Leuven, Radiology Department, Leuven, Belgium 7 Aarhus University Hospital, Medical Physics Department, Aarhus, Denmark 8 Aarhus University Hospital, Oncology Department, Aarhus, Denmark 9 Aarhus University Hospital, Pathology Department, Aarhus, Denmark Abstract text The Gleason score (GS), a well-validated prognostic factor, is used for risk stratification and treatment selection in many nomograms and estimates the aggressiveness of prostate cancer. The GS is an aggregate of the first and second most dominant Gleason Patterns (GP). Usually, random biopsies bear the risk of underestimating the extent of the disease and missing high-grade GP compared to whole-mount section pathology after prostatectomy. This is a key motivation for targeted biopsies, either using fused MR-ultrasound guidance, or direct MR guidance. Where several studies aimed to predict the aggregate GS from MRI on patient- or tumor-level, we sought spatial mapping of high Gleason grade components inside the lesions on voxel- level using quantitative mp-MRI (qMRI) to improve the accuracy of targeted biopsies for sampling high-grade GP within heterogeneous prostate tumors. For this purpose in a multicenter setting, we used individual patient- customized 3D-printed molds to register the high resolution (0.5 µm/pixel) segmentations of GP 3, 4, and 5

Symposium: Why is fully automated image segmentation and deformable image registration not here yet?

Made with FlippingBook - Online magazine maker