Abstract Book

ESTRO 37

S420

Results A total of 138 patients with pretreatment biopsies who completed planned chemo radiation were included. The median EQD2 to point A was 81 Gy (68-84Gy) and 85% patients received concurrent cisplatin. Overall 62 (45%) patients had stage IB2-IIB and 76 (55%) patients had stage III-IVA tumors. Of these 34 had pelvic nodes (24.6%) and 9 had paraortic nodes (6.5%). The median tumour volume was 262 cc (IQR 98-380cc). SOX-2, OCT 4,Nanog, CD44, Podoplanin expression was seen in 36.2%, 25%, 35.3%, 30% and 8.5% of patient samples. Corresponding score ≥ 2+ was observed in 8.6%, 5.7%, 26%, 26% and 1.7% samples. At a median follow up of 27 mths, local relapse was observed in 17/138 patients (12.3%) and distant in 31/138 (22.4%) patients with overall relapse rate of 30.6%. Three year local relapse free survival was higher in patients with ESC+/CD44+ tumours as compared to ESC+/CD44- tumours (100% vs. 55.5%p=0.02). Superior 3-year DFS was observed in ESC+/CD44+ cohort as compared to ESC+/CD44-ve cohort(60% vs 37%, p=0.06). Pelvic nodal involvement (62.8% vs 31.4%, p=0.07), paraortic nodal involvement (62% vs 0%, p=0.001) and advanced stage >IIB predicted for determent in 3 year DFS (71% vs 48% p=0.02). On multivariate analysis stage >IIB (p=0.02), paraortic nodes (0.02) and ESC+/CD44- cell population (0.005) independently predicted for reduced DFS. Conclusion Loss of CD44 in ESC expressing cells is associated with reduced DFS as compared to ESC expressing cells that retain CD44 expression. As loss of CD44 is linked with increased metastasis further studies investigating ESC, CD44 and epithelial mesenchymal pathways are warranted. PO-0807 Diversity of PET imaging biomarkers predicting cervical cancer treatment outcome: where do we stand? S. Reuzé 1,2,3 , C. Chargari 2,3,4,5,6 , A. Schernberg 3,6 , R.D. Seban 7 , A. Alexis 1,3 , S. Bockel 6 , K. Berthelot 6 , A. Escande 6 , L. Dercle 7,8 , C. Haie-Meder 6 , E. Deutsch 2,3,6 , C. Robert 1,2,3 1 Gustave Roussy, Radiotherapy Department- Medical Physics Unit, Villejuif, France 2 INSERM U1030, Molecular Radiotherapy, Villejuif, France 3 Paris-Saclay University, Faculty of Medicine, Le Kremlin- Bicêtre, France 4 French Military Health Services Academy, Ecole du Val- de-Grâce, Paris, France 5 Institut de Recherche Biomédicale des Armées, Paris, France 6 Gustave Roussy, Radiotherapy Department, Villejuif, France 7 Gustave Roussy, Nuclear Medicine Department, Villejuif, France 8 INSERM, U1015, Villejuif, France Purpose or Objective With the advent of high performance computing methods allowing more quantitative medical imaging analysis, a large amount of candidate image-based biomarkers are investigated to elaborate predictive signatures of treatment outcome, or distinguish between subtypes of cancer. Radiomics, a method based on the extraction of numerous intensity, shape, or textural features can lead to thousands of indices from all imaging modalities. Treatment of locally advanced cervical cancer (LACC) involves a large amount of multimodal images (CT, MRI, PET) at several steps of the treatment: diagnosis, radiotherapy planning, brachytherapy (BT) planning, follow-up. At each step, imaging biomarkers can be extracted to predict survival or recurrence and eventually adapt the next treatment step. Several image biomarkers (tumor size, volume, radiotracer uptake) are already reported in clinical reports, whereas radiomics

analysis needs a complementary step using a specific software. Discriminative power of biomarkers depends of tumor type and modality. No consensus was proposed in favor of a specific biomarker in the literature. This study aims to sum up institutional findings about imaging biomarkers of LACC relapse. Material and Methods 118 patients were retrospectively included in this threefold study. All of them underwent homogeneous chemoradiation and image-guided pulsed-dose-rate BT treatments. A radiomic analysis based on a threshold- based tumor segmentation, image preprocessing, feature extraction and statistical analysis was performed on baseline PET images to establish a signature of local recurrence. In parallel, a score combining both a PET quantitative index from the tumor and neutrophil count from biological reports was defined to evaluate local control (LC). Finally, a score based on FDG uptake in bone marrow (BM) was defined to evaluate local survival (LS) and local recurrence free survival (LRFS). For all analyses, cohort was separated in two subsets with different image acquisition parameters to permit training and validation of the model. Results A 4-feature PET-based radiomic signature of local recurrence was established (training set: AUC=0.86; validation set: AUC=0.76, p<0.001). A score combining neutrophilia and SUVpeak of the tumor measured on PET images was associated with worse local control (training set: HR=9.2, p<0.001; validation set: HR=3.3, p=0.03). Finally, a high BM SUVmax was associated with lower LS and LRFS (HR=4.3, p<0.0001; HR=4.3, p=0.001 respectively). Conclusion All these studies based on routinely collected data showed promising results while approaches were different (Fig.1). This suggested that baseline FDG-PET acquisitions were highly predictive of treatment outcome. Radiomics studies usually involve a huge processing pipeline whereas conventional biomarkers are more commonly reported by physicians, and more accessible for retrospective analysis. The automatisation of radiomics in manufacturer softwares could be an important step for a clinical use.

PO-0808 Efficacy and toxicity of chemoradiation with brachytherapy for locally advanced cervical cancer N. Horeweg 1 , C.L. Creutzberg 1 , E.C. Rijkmans 1 , M.S. Laman 1 , L.A. Velema 1 , V.L.M.A. Coen 2 , T.C. Stam 3 , E.M. Kerkhof 1 , J.R. Kroep 4 , C.D. De Kroon 5 , R.A. Nout 1 1 Leiden University Medical Center LUMC, Department of Radiotherapy, Leiden, The Netherlands 2 Zuidwest Radiotherapy Institute, Department of Radiotherapy, Vlissingen, The Netherlands 3 Medical Center Haaglanden, Department of Radiotherapy, Leidschendam, The Netherlands 4 Leiden University Medical Center LUMC, Department of Medical Oncology, Leiden, The Netherlands 5 Leiden University Medical Center LUMC, Department of Gynaecology, Leiden, The Netherlands