Abstract Book

S92

ESTRO 37

model incorporating multiparametric MRI (mpMRI). The model requires knowledge of tumour location, tumour cell density, tumour aggressiveness and hypoxia. In this work we describe a quantitative, voxel-by-voxel approach for extracting tumour characteristics from mpMRI. Material and Methods Forty–five patients scheduled for radical prostatectomy underwent in vivo mpMRI. In addition, sub-groups of patients underwent BOLD MR and PSMA-PET scanning. Quantitative parametric and pharmacokinetic maps, including ADC, Ktrans, Kep, ve and R2* were generated. The surgical specimen was embedded in a purpose-built sectioning box for accurate 3D co-registration with mpMRI data. Digital histology data was processed to generate cell density maps and define regions of low/high grade disease. These data, with pathologist annotations, were co-registered using a previously described sophisticated framework (3). Histology stained with 3 hypoxia biomarkers (HIF1α, CAIX and GLUT1) was processed to generate expression levels for correlation with the mpMRI data. Advanced statistical and machine learning techniques were used to generate mathematical models to predict tumour location, cell density tumour grade and presence of hypoxia. Results Prediction of tumour location using support vector machines has demonstrated a prediction accuracy ranging from 70.4 to 87.1%. Haralick’s texture features were computed from mpMRI in 29 patients. Two regression methods demonstrated promising results for cell density prediction. Logistic regression models demonstrated accuracies >83% in differentiating between high and low grade tumours. Regarding hypoxia, HIF1α and CAIX expression levels demonstrated significant correlations with pharmacokinetic maps generated from DCE (e.g. for HIF1α and: k ep p =0.01 and ve p= 0.04). Additional validation of these results using next generation sequencing is under way. Conclusion Parametric and pharmacokinetic maps generated from mpMRI and a sophisticated framework for co-registration of ground truth histology with mpMRI has provided a platform to develop predictive models for tumour location and biology. OC-0176 Tumor volume and diffusivity during RT show non-intuitive interrelation with treatment outcome F. Mahmood 1 , H.H. Johannesen 2 , P. Geertsen 1 , R.H. Hansen 2 1 Herlev and Gentofte Hospitals, Radiotherapy Research Unit- Department of Oncology, Herlev, Denmark 2 Herlev and Gentofte Hospitals, Department of Radiology, Herlev, Denmark Purpose or Objective Solid tumor response is generally evaluated using size criteria. Unfortunately, size change manifests slowly and the treatment outcome cannot be evaluated sooner than weeks/months after treatment. In this study the temporal evolution of the size and diffusivity of brain metastases during fractionated radiotherapy (RT) was monitored using repeated magnetic resonance imaging (MRI). The main aim was to investigate their relation to Twenty-nine metastases (N=29) from twenty-one patients were analyzed prospectively. Patients were scheduled for 30 Gy in ten fractions. A 1 T MRI system was used to acquire DW-, T2W- and CE T1W-MRI, before RT and at follow-up 2-3 months after RT. Additionally, DW-, T2W- the treatment outcome. Material and Methods Proffered Papers: PH 3: Functional imaging

scans were acquired within 1 hour before or after each RT fraction. Volume estimation during RT was based on T2W-MRI. The diffusivity regions of the viable tumor were outlined using b =800 s/mm 2 DW-MRI and the apparent diffusion coefficient (ADC) was estimated using four b - values ( b =400-800 s/mm 2 ) in a mono-exponential model. Treatment outcome was evaluated by volume criteria using T1W scans and thresholds from RECIST. The prognostic capacity of ADC was evaluated by the area under the curve (AUC) of receiver operating curves using Matlab (α=0.05, H1: AUC > 0.5). Results Non-responding (progressive disease (PD) or stable disease (SD)) metastases (n=10) showed no mean volume change during the first six fractions and a mean reduction of about 25 % during the remaining 4 fractions, stabilizing at fraction 9 (Fig. 1a). Responding (partial response (PR) and complete remission (CR)) metastases (n=19) showed a large variation in volume change and no overall trends (Fig. 1b). The overall composition of the tumor i.e. volume fraction of viable tumor remained constant regardless of treatment outcome (data not shown). The mean ADC increased almost monotonously in responding metastases, and decreased in nonresponding metastases, leading to a statistical significant capacity to stratify treatment response at fraction 3, and more permanently from fraction 7 and on (Fig. 2a). Further, neither tumor histology nor baseline tumor volume was correlated to treatment outcome (Fig. 2b).

Fig. 1 Relative volume change in a. Non-responding metastases, b. responding metastases. Error bar = 1 SD.