ESTRO 37 Abstract book

ESTRO 37

S537

6 The Christie NHS Foundation Trust, Department of Radiotherapy Related Research, Manchester, United Kingdom Purpose or Objective An accurate estimation of lung tumour volume from [18F]-fluorodeoxyglucose positron emission tomography- computed tomography (FDG PET-CT) is important for radiotherapy planning and therapy response assessment. Segmentation based on the standard uptake value (SUV) is often used for this purpose. The most common segmentation method is based on the SUVmax, whereby the tumour border is defined by the contour curve at a percentage threshold (ranging from 40-50%) of the maximum intensity voxel within a predefined region of interest. However, since this maximum reflects the value on a single pixel, it may be affected by noise in the image leading to an inaccurate segmentation. The SUVpeak has been proposed as an alternative evaluation method, whereby a uniform averaging filter with a 1 cc kernel is used to evaluate the SUV. Since this method does not rely on a single pixel, it is less likely to be affected by noise and may therefore be a more reliable value to base the segmentation on. The aim of the study was to compare simple SUVpeak and SUVmax based segmentation methods for calculating the lung tumour volume, compared to a pathology ground truth. Material and Methods Thirty patients diagnosed with stage 1 and 2, Non-Small Cell lung cancer underwent surgical resection in the Netherlands between 2006 and 2008. All pathology samples were macroscopically and microscopically analysed and tumour volumes calculated. FDG PET-CT scans for these patients were acquired before surgery (median 20 days and range 1 to 57 days). The tumour volume found at each percentage threshold from the SUVmax and SUVpeak, were generated for each patient. The percentage threshold that provided the tumour volume that corresponded with the pathology volume was considered to be the optimal threshold. Using fits for these optimal thresholds as function of tumour volume (v) all tumours were segmented and equivalent diameter (d) was calculated using the formula d=2(3V/4π) 1/3 . This was then compared with the equivalent diameter of the pathology volume. Results The mean optimal percentage threshold was (82% +/-31%) and (48% +/-10%) for the SUVpeak and SUVmax respectively, where the optimal threshold increased strongly with smaller tumour sizes (Fig.1). The optimal threshold could be fitted well with power law function (Fig.1), with a higher correlation for the optimal percentage threshold for the SUVpeak (R 2 =0.70) as opposed to the SUVmax (R 2 =0.18). After iterative segmentation to account for the volume dependent threshold, the mean pathology equivalent diameter ratios were (0.93 +/-0.21) and (0.96 +/-0.16) for the SUVpeak and SUVmax respectively (Fig.2).

improved the performance to an AUC of 0.90 (Model 2 Table). Parotid gland dose did not significantly add to the ΔIBM models (log-likelihood ratio test; p>0.06).

Conclusion The ΔIBM parotid gland surface change (ΔSurface) between the 3rd week during and start of treatment was associated with the development of XER12m. The ΔIBM model with ΔSurface and XERbaseline improved the performance (AUC = 0.87) with respect to the reference model (AUC = 0.84). Addition of XERw3 further improved the performance of this ΔIBM model (AUC = 0.90). PO-0973 SUVpeak based segmentation to determine lung tumour volume on FDG PET-CT compared with pathology S. Mercieca 1 , J. Belderbos 2 , J. Van Loon 3 , K. Gilhuijs 4 , P. Julyan 5 , M. Van Herk 6 1 University of Malta, Department of Radiography, Msida, Malta 2 The Netherlands Cancer Institute Antoni van Leeuwenhoek Hospital, Radiotherapy Department, Amsterdam, The Netherlands 3 Maastro Clinic-, Department of Radiation Oncology, Maastricht, The Netherlands 4 University Medical Centre, Department of Medical Physics, Utrecht, The Netherlands 5 The Christie NHS Foundation Trust, Nuclear Medicine Department, Manchester, United Kingdom