ESTRO 38 Abstract book

S1047 ESTRO 38

Material and Methods The duration of significant mucositis in 81 patients with oropharyngeal cancer was measured via four outcome metrics (CTCAE versions 3 and 4 (grade 3 or greater), patient reported duration of interference in daily activities (quite a bit or very much) and patient reported duration of severity (severe or very severe)). The oral mucosa volume was outlined according to international consensus guidelines. The section of the dose grid intersecting the oral mucosa was extracted and converted to BED. Thirty nine intensity and spatial metrics were extracted from the dose grid using Pyradiomics ( van Griethuysen, J. J. M et al, 2017) . Varimax rotated PCA was used to de-correlate the data and reduce the thirty nine dimensions down to the ten dimensions with the greatest variance. These ten variables were combined into a single model predicting the duration of toxicity. The dose bin width for the calculation of dose metrics was set to be 1 Gy 10 BED wide. Results The median duration was 7 weeks for all toxicities except for patient reported severity which was 6 weeks. The interquartile range was 4-5 weeks with outliers ranging from 0 to 15 weeks. When correlating the principal components against toxicity duration two components remained statistically significant after applying a Benjamini-Hochberg correction namely CTCAE V4 and the patient reported duration of severity outcomes. These two features were strongly anti-correlated although the dominant features in the components were different. The first component was dominated by metrics related to the spread of doses such as variance. The other component was dominated by the median and image coarseness, a measure of the rate of change of the dose. A third component was also significant until the multiple comparison correction. Lasso regression identified similar metrics to be important as those found by varimax rotated PCA analysis. Conclusion Spatial metrics were not found to correlate any more strongly to the duration of mucositis than simpler metrics. The most strongly correlated factors were metrics such as median dose and variance rather than the spatial metrics. Factor analysis suggested metrics related to the variance in the dose delivered to the oral mucosa and the rate of change in dose within a region was correlated to duration of toxicity. Outcome modelling with regression and classification is difficult with no strong classifiers. Work is ongoing to try and create a model to identify patients at risk of long toxicity duration. EP-1925 Association of MRI-based radiomic features with prognostic factors in oropharyngeal cancer G. Marvaso 1 , C. Delia 1 , D. Alterio 1 , F. Botta 2 , C. Giannitto 3 , S. Volpe 1,4 , F.A. Maffini 5 , S. Raimondi 6 , M. Radiotherapy, Milan, Italy ; 2 IEO- European Institute of Oncology IRCCS, Unit of Medical Physics, Milan, Italy ; 3 IEO- European Institute of Oncology IRCCS, Radiology, Milan, Italy ; 4 University of Milan, Department of Oncology and Hemato-oncology, Milan, Italy ; 5 IEO- European Institute of Oncology IRCCS, Division of Pathology, Milan, Italy ; 6 IEO- European Institute of Oncology IRCCS, Epidemiology and Biostatistics, Milan, Italy ; 7 IEO- European Institute of Oncology IRCCS, Otolaryngology and Head and Neck Surgery, Milan, Italy Purpose or Objective Radiomics focuses on extracting a large number of quantitative imaging features correlated with clinical characteristics. We propose a radiomic approach using magnetic resonance imaging (MRI) to decode tumour phenotype and treatment response in oropharyngeal squamous cell carcinoma (OPSCC). Ansarin 7 , M. Bellomi 3 , B.A. Jereczek-Fossa 1,4 1 IEO- European Institute of Oncology IRCCS,

Results Correlation analysis on the whole dataset highlights presence of several disjoint clusters made up of many highly correlated RFs, see Fig. 2A . Intra-scanner analysis, Fig. 2B-C, also revealed similar structures. Using a threshold of 0.85, 27 and 25 nonredundant RFs were selected for ROIg, considering, separately, data acquired with scanner A and B. The selected RFs were 27 and 30 for ROIc, 26 and 30 for ROIb. Strong clustering coherence was observed for GLRLM25 features (all the examined ROIs have the same class label set in common). Also, GOH RFs have a higher number of stable clusters in common among all the ROIs compared to GLCM25. Likewise, preliminary intra-patient analysis confirmed the same clustering coherence for GLCM25 and GOH features.

Conclusion The proposed methodology of clustering coherence analysis allowed to reduce to about 1\10 the number of RFs to be used for further analyses. A-priori reduction of RF number will enhance generalization capability of radiomic signatures and speed up implementation of radiomic approaches into clinical practice. EP-1924 Are spatial dose metrics predictive of oral mucositis duration in oropharyngeal cancer? A. Carver 1 , M. Hickman 2 , S. Warren 1 , M. Ward 1 , A. Thomas 2 , P. Sanghera 3 , C. Fong 3 , A. Hartley 3 1 University Hospital Birmingham NHS Foundation Trust, Medical Physics, Birmingham, United Kingdom ; 2 University Hospital Birmingham NHS Foundation Trust, Radiotherapy, Birmingham, United Kingdom ; 3 University Hospital Birmingham NHS Foundation Trust, Clinical Oncology, Birmingham, United Kingdom Purpose or Objective In patients with oropharyngeal cancer treated with chemoradiotherapy oral mucositis is very likely to occur at a significant grade and the pertinent question is the duration rather than the probability of occurrence. It has been established that simple dosimetric parameters do not correlate well to the duration of toxicity in these patients (Hickman, M. 2017). This study uses spatial dose metrics to try to establish whether there is a more complex relationship between dose and toxicity duration.