ESTRO 38 Abstract book

S1046 ESTRO 38

cohorts, highlighting the importance of daily imaging and the use of EQD d rather than D a . The impact of EQD d on NTCP outcome is being investigated. Building on the results presented, the study of pelvic and H&N cases is being extended, exploiting the 194 available H&N cases as well as a dataset of 250 pelvic cases from the VoxTox programme. For local analysis, dose surface maps of the rectum are obtained for both D a and EQD d , and are compared for dose-response modelling. EP-1923 Dimensionality reduction of radiomic features using a clustering coherence-based approach C. Tenconi 1,2 , T. Rancati 3 , F. Palorini 3 , A. Cerrotta 4 , B. Pappalardi 4 , F. Piccolo 4 , A. Messina 5 , M. Carrara 6 , T. Giandini 6 , C. Fallai 4 , E. Pignoli 6 , L. Licitra 1,7 , R. Valdagni 1,2,3 1 University of Milan, Department of Oncology and Haemato-oncology, Milan, Italy ; 2 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiation Oncology 1, Milan, Italy ; 3 Fondazione IRCCS Istituto Nazionale dei Tumori, Prostate Cancer Program, Milan, Italy ; 4 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiation Oncology 2, Milan, Italy ; 5 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiology, Milan, Italy ; 6 Fondazione IRCCS Istituto Nazionale dei Tumori, Medical Physics, Milan, Italy ; 7 Fondazione IRCCS Istituto Nazionale dei Tumori, Head and Neck Medical Oncology Unit, Milan, Italy Purpose or Objective Radiomic features (RFs) dimensionality represents a critical challenge to be addressed to lower the overfitting risk in radiomic studies. Assess a methodology for a-priori selection of non-redundant RFs from MR acquisitions and investigate its robustness over different clinical scenarios. Material and Methods T2-weighted (T2W) images of 20 patients (pts) with locally advanced cervical cancer, treated with 4 session MR- guided adaptive brachytherapy were considered. Within ~12days, all pts underwent multiple MR with a standardized acquisition protocol. Fig. 1 illustrates study workflow: 77scans acquired with 2 different 1.5T scanners (same manufacturer) were uploaded to Image Biomarker EXplorer (IBEX) for RF extraction. Cylindrical regions of interest (ROIs) were manually drawn on 3 different anatomical regions, possibly representing 3 different radiomic scenarios: the left gluteus muscle (ROIg), outside treatment region (variation in RFs due to factors other than treatment); the cervix tumor target volume (ROIc), (variation in RFs induced by treatment); the bladder (ROIb), which is usually filled before each scan (uniform texture). From these ROIs 279 RFs were extracted for Gray Level Cooccurrence Matrix 25 (GLCM25) , Gray Level Run Length Matrix 25 (GLRLM25) , Gradient Orient Histogram (GOH) categories. To explore linear associations among RFs and to investigate clustering coherence under different clinical/technical scenarios (inter-scanners, intra-patient, inter-ROIs), correlation matrices were computed for the whole dataset and for data grouped according to MR scanner, to ROI type (i.e.: ROIg/ROIc/ROIb) and to acquisition session. For each correlation matrix, clusters of highly correlated RFs (threshold r=0.85) were identified and a “class label” assigned to the RF with the simplest mathematical definition. Redundant RFs (i.e., RFs with similar definition) were filtered out from the dataset for further analyses.

1 German Cancer Research Center DKFZ, Department of Medical Physics in Radiation Oncology, Heidelberg, Germany ; 2 Heidelberg University, Department of Physics and Astronomy, Heidelberg, Germany ; 3 National Center for Radiation Research in Oncology NCRO, Heidelberg Institute for Radiooncology HIRO, Heidelberg, Germany ; 4 German Cancer Research Center DKFZ, Clinical Cooperation Unit "Radiation Oncology", Heidelberg, Germany ; 5 University Hospital Heidelberg, Department of Radiation Oncology, Heidelberg, Germany ; 6 University of Cambridge, Cavendish Laboratory, Cambridge, United Kingdom ; 7 Cambridge University Hospitals NHS Foundation Trust, Department of Oncology, Cambridge, United Kingdom ; 8 University of Cambridge, Department of Oncology, Cambridge, United Kingdom ; 9 Freiburg University Medical Center, Department of Radiation Oncology, Freiburg, Germany ; 10 University of Cambridge, Department of Engineering, Cambridge, United Kingdom ; 11 Cambridge University Hospitals NHS Foundation Trust, Department of Medical Physics and Clinical Engineering, Cambridge, United Kingdom ; 12 German Cancer Research Center DKFZ, Medical Physics in Radiation Oncology, Heidelberg, Germany Purpose or Objective Daily imaging facilitates investigation of the impact of anatomical changes on the delivered dose. Conventional dose accumulation averages the fraction dose which introduces a systematic mathematical underestimate of the achieved biological effect in the presence of dose variations. To overcome this inconsistency in the use of the total delivered dose for biological models, we introduce an EQD-based formalism to accumulate the total biological dose. The difference with respect to conventional dose accumulation is investigated for two different patient cohorts using daily imaging. Material and Methods We adapted the standard EQD-formalism to compare a treatment of constant fraction dose d to a treatment of daily varying doses d i . The derived EQD d (d i ) yields the total biological dose under consideration of the varying delivered doses and the tissue radiosensitivity. An initial cohort of 9 patients treated for prostate carcinoma with IMRT was studied using daily imaging with an in-room CT- on-rails system. Gamma index analysis (3%/3mm) is used to compare both the conventionally accumulated (D a ) and the biologically accumulated (EQD d ) with the planned dose. The second test cohort consisted of 30 out of 194 available patients treated for head and neck (H&N) tumours recruited in the VoxTox programme that underwent TomoTherapy and daily MVCT imaging. Results EQD d is systematically higher than D a with highest deviations in dose gradient regions around the target and in areas of strong motion amplitudes. In the pelvic cases, deviation hot spots in the bladder and rectum wall around 4 Gy were found for 5/9 patients. In 6/9 cases for both bladder and rectum using EQD d , 2-4 Gy are added up locally in regions where the gamma criterion already failed using D a , and furthermore leading to a failure rate increase of up to 3%. H&N patients showed an overall smaller difference between the two accumulation methods. Deviations are below 1 Gy for most volumes but deviation hotspots between 1 Gy and 8 Gy in OARs around the target volume were found in 20% of cases. Conclusion The systematic underestimation of the biological effect from dose accumulation can potentially impair dose- response modelling and treatment adaptation. The presented approach of biological dose accumulation can avoid this inaccuracy. Stronger day-to-day motion amplitudes in the pelvic region compared with the H&N cases resulted in higher deviations of several Gy. Individual cases of high local deviation occurred in both