Three Years of RTDS: A British Success Story

S622

International Journal of Radiation Oncology  Biology  Physics

studies, CTCAE criteria have been used more often than other criteria (86.4%, 57/66). Conclusions: Several scoring systems exist today to assess RT related toxicity. The use of different evaluation standards has made comparison of therapy effects across studies difficult. This review discusses the commonly used criteria, such as CTCAE, RTOG\EORTC, LENT-SOMA scoring criteria, and introduces preliminary assessment of the use of three scoring criteria in the recent three years. The CTC\CTCAE criteria are becoming the most commonly used standard, although it is still not perfect. Sometimes it is used in combination with other scoring systems. Finally, even though CTC\CTCAE use is on the rise, the RTOG\EORTC scoring criteria are still in use in nearly half of the studies. Author Disclosure: Y. Jiang: None. L. Yuan: E. Research Grant; Varian. Q. Wu: E. Research Grant; Varian, NIH. F. Yin: E. Research Grant; Varian. Y. Ge: E. Research Grant; NIH.

3096

3095 Malignant Glioma Delineation in Amino Acid PET-Images Using a 3D Random Walk Approach T. Fechter,1 M. Mix,2 I. Gardin,3 T. Papke,1 T. Schimek-Jasch,1 P. Meyer,2 A. Grosu,1 and U. Nestle1; 1Department of Radiation Oncology, University Hospital Freiburg, Freiburg, Germany, 2Department of Nuclear Medicine, University Hospital Freiburg, Freiburg, Germany, 3Centre Henri Becquerel, Rouen, France Purpose/Objective(s): The delineation of malignant gliomas presents a frequent problem in the field of radiation therapy as they are often not well depicted by CT and MRI. An alternative presently clinically investigated is amino acid (e.g., FET) PET imaging with higher diagnostic accuracy as compared to morphological imaging. However, most common PET image segmentation approaches fail to segment the glioma accurately due to its heterogeneity and the rather low contrast. A possible alternative to the common, threshold based approaches are probability theory algorithms like the random walk (RW). Previous RW contouring studies showed promising results with inhomogeneous tumors and images with weak boundaries. Materials/Methods: The presented RW approach pre-segments a user defined region of interest into 3 classes (tumor, background and unlabeled pixels) based on the pixel intensities. Every unlabeled pixel is then assigned to the class which is most likely a random walker would reach first starting from that pixel. The walker is driven by the images gradient field and pixel intensities. 14 amino acid PET datasets with a glioma were contoured. The RW results were compared to the outcomes of a 40% threshold and a 1.4 opposite mean (OM) threshold method plus the GTV delineated by a clinician. Results: Clinically plausible contours were produced in 6/14 cases by the 40% method, 3/14 cases by the OM method and 0/14 cases by the RW method. In cases with usable contours, false positive spots outside the tumor region were seen in 5/8 for the 40% method, in 10/11 for the OM method and in 0/14 for the RW method being a sign for robustness against noise. However, in 4 cases the RW method delineated a rather big area in longitudinal direction. In cases with usable contours produced by all methods, the mean volume was 47.03 mL for the 40%, 54.70 mL for the OM, 30.20 mL for the RW method and 40.12 mL for the GTVs Conclusions: The presented work suggests that the RW approach may be a robust method for the segmentation of low contrast PET images. Here, the method outperforms common threshold based segmentation techniques by applicability and robustness against noise. Further research will explore the possible use of this method in clinical practice. Author Disclosure: T. Fechter: None. M. Mix: None. I. Gardin: None. T. Papke: None. T. Schimek-Jasch: None. P. Meyer: None. A. Grosu: None. U. Nestle: None.

Multi-institutional/Technique Dose Comparisons Using Overlap Volume Histograms for Difficulty Normalization J. Moore,1 A.S. Dholakia,1 E.T. Kimberly,1 L. Wang,2 A. Koong,2 K. Goodman,3 J.M. Herman,1 and T.R. McNutt1; 1Johns Hopkins University, Baltimore, MD, 2Stanford Comprehensive Cancer Center, Stanford, CA, 3Memorial Sloan-Kettering Cancer Center, New York, NY Purpose/Objective(s): With rapidly evolving radiation therapy techniques, we have yet to fully understand the advantage in terms of dose distribution for each approach compared to the others. This study compares dose distributions from two different delivery techniques and from different institutions by using overlap volume histograms (OVHs) to normalize the plan difficulty of different patients. The OVH normalization allows for comparison of dose attributes of clinically treated patient plans while removing the influence of structure proximity. Materials/Methods: A SQL database containing 53 pancreas stereotactic body radiation therapy (SBRT) patients is populated with dose and structure relationship information. The patient population consists of patients from three separate institutions treated according to the objectives specified in the treatment protocol and contains dose information for 39 IMRT plans and 6 Dynamic Arc patients. A script in the treatment planning system is used to automatically compute OVHs and insert both OVHs and dose volume histograms (DVHs) into the database. The DVHs and OVHs of each patient are queried at the volume points specified in the treatment protocol. To normalize the plan difficulty between patients, for each volume point of interest, the dose and distance are plotted. Trends are compared between the plots of each group. Results: Thirty-one of the patients had both dose and structure information for the duodenum available in the database (25 IMRT, 6 dynamic arc). When normalized by overlap distance, dose to the duodenum at 1 cc trends lower for dynamic arc fields when compared to the IMRT. Dose at 1 cc and 9 cc is lower on average for dynamic arc compared to IMRT (p Z 0.01, 0.04 respectively) while the average overlap distance is not significantly different between the two groups (p Z 0.40, 0.50 respectively). At 3 cc, dose is lower for dynamic arc plans, though the overlap distance is greater. The achieved dose for patients with overlapping structures show tradeoffs with distance, while dose decreases with distance for patients with structures that do not overlap. Furthermore, dynamic arc patients received significantly reduced dose (p < 0.01) to the kidneys even though they had significantly shorter distance to overlap than IMRT patients (p Z 0.02). Conclusions: Dynamic arc planning for SBRT patients has the potential to reduce dose to normal structures even for patients with shorter overlap distances. Using OVHs in comparisons allows plan difficulty to be separated from achieved dose values and allows for plan quality to be compared across institutions with actual clinical plans used for treatment. Author Disclosure: J. Moore: E. Research Grant; Philips. A.S. Dholakia: None. E.T. Kimberly: E. Research Grant; Elekta. L. Wang: None. A. Koong: None. K. Goodman: None. J.M. Herman: None. T.R. McNutt: E. Research Grant; Philips, Elekta. L. Stock Options; Philips. O. Patent/ License Fee/Copyright; Accuray, Elekta.

3097 Three Years of RTDS: A British Success Story C. Ball; Clatterbridge Cancer Centre NHS FT, Wirral, United Kingdom Purpose/Objective(s): To describe the experience of initiating a National Dataset for radiation therapy (RTDS) within the United Kingdom. Materials/Methods: When the radiation therapy Ko¨rner return was disbanded in 1995, there was no longer a central collection of data for radiation therapy activity. The National Cancer Services Analysis Team (NatCanSat) formed in 1998 by Dr Brian Cottier recognized the gap in information and looked towards the Oncology Management System

Volume 87  Number 2S  Supplement 2013

Poster Viewing Abstracts S623

Poster Viewing Abstract 3097; Table 2009/2010

2010/2011

2011/2012

Episodes Attendances Episodes Attendances Episodes Attendances Total Male Female Breast Urology Head and neck Lung

121,289 56,847 63,678 34,144 21,996 6098

1,647,249 803,001 843,638 485,611 376,994 142,837

125,506 58,806 66,404 35,589 23,158 6395

1,719,607 849,608 869,407 506,676 413,363 148,913

130,420 61,653 68,728 37,132 25,366 6464

1,793,552 904,125 889,029 510,076 459,117 153,007

16,394

124,599

16,985

132,214

17,736

141,890

(OMS) which holds a wealth of electronic information on radiation therapy treatment. A pilot project investigated the feasibility of extracting data from the OMS and linking it with patient demographic data found only in the hospital administration systems. It became apparent the processes of data entry and coding with the OMS were inconsistent and for central reporting to become feasible a common set of currencies, formatting and coding need to be adopted. NatCanSAT took the first step towards standardization by creating electronic toolkits which extracted ONS data, imported it and processed it into a standard return. NHS Trusts have the ability to quality assure and report information locally before submitting it. The toolkits are available to all radiation therapy providers in the UK. It allows the user to run an extract without knowledge of radiation therapy and/or database structures. Since April 2009, NatCanSAT accepts records on every patient treated with radiation therapy funded by the National Health Service (NHS) in the UK. Results: The aggregated data is reflected back to 55 radiation therapy centers and commissioners via a secure dedicated website, but the RTDS also provides unprecedented opportunities to examine the overall picture of radiation therapy, and to compare practice in many ways. The summary Table shows the number of records submitted for England. Conclusions: RTDS has had many positive spin-offs: 1) Agreed currencies and definitions; 2) A convergence of working practices in the way OMS are used, which should improve safety; 3) Better understanding of their data by radiation therapy, management staff and commissioners; and 4) The ability to link patient data to geography can provide evidence to assist planning the location of radiation therapy services. Beyond driving the financial and management arrangements for radiation therapy services, the RTDS provides rich opportunities for clinical investigations and research which can help to develop practice. Author Disclosure: C. Ball: None.

3098 Applicability of a General Predictive DVH (pDVH) Model to Rare Treatment Sites With No Prior Training Data L. Appenzoller,1 J.R. Olsen,1 S. Mutic,1 and K.L. Moore2; 1Washington University School of Medicine, Saint Louis, MO, 2University of California, San Diego, San Diego, CA Purpose/Objective(s): Knowledge-based treatment planning for automation and the reduction of IMRT plan quality variation is a topic of increasing interest in clinical radiation therapy. All current methods to predict achievable OAR DVHs (pDVH) require prior patient training data that may not be available for infrequently treated sites. The objective of this work was to investigate the application of a general pDVH model trained from a common treatment site (rectum OAR DVHs in prostate cancer) to the prediction of OAR DVHs in various infrequent or unique treatment sites. Materials/Methods: Five patients treated with IMRT to uncommon disease sites with a high priority OAR in close proximity to the PTV were selected for this study. The treatment site/OAR pairs included thigh sarcoma/femur, pancreas/duodenum, retroperitoneal sarcoma/bowel, retreatment rectum/bowel, and psoas/cauda equina, respectively. The five

clinically approved IMRT plans were replanned to ensure maximum sparing of OARs and preservation of PTV quality metrics in agreement with site specific clinical objectives. A previously validated pDVH model to predict achievable rectum OAR DVHs for prostate cancer patients derived based on the correlation of expected dose to the distance from a voxel to the PTV surface was used to make a DVH prediction for the 5 replanned OARs of interest. A sum of residuals (SR) metric quantified the integrated difference between the DVH and pDVH and was used to evaluate the ability of the rectum pDVH model to accurately predict achievable OAR DVHs for these 5 unique cases. The SR metric was also used to assess the improvement in OAR sparing in the replanned cases. Results: The accuracy of the rectum pDVH model in the prediction of various OAR DVHs in unique treatment sites was demonstrated in all five cases with the SR values near zero and within one standard deviation of the average SR values in the prostate pDVH model validation cohort (SR Z 0.003 +/- 0.037). There was a large reduction in SR values between the clinically approved plan and the replan in 3 out of 5 cases: thigh sarcoma (0.1030 to 0.0300), retroperitoneal sarcoma (0.1050 to 0.0008), and psoas (0.1701 to 0.0170) patients. This result implies that using the rectum pDVH model’s standard deviation as a quality threshold would have correctly identified plans with unrealized OAR sparing. Conclusions: The results of this study suggest that a general pDVH model trained with patients from a commonly treated site can have predictive value in various unique treatment sites with minimal or no prior training data. This feature may be useful for providing patient-specific guidance in clinical situations where minimal training data is available and attainable clinical objectives are uncertain. Author Disclosure: L. Appenzoller: O. Patent/License Fee/Copyright; Developing Predictive Dose-Volume Relationships for a Radiation therapy Treatment. J.R. Olsen: None. S. Mutic: O. Patent/License Fee/Copyright; Developing Predictive Dose-Volume Relationships for a Radiation therapy Treatment. K.L. Moore: O. Patent/License Fee/Copyright; Developing Predictive Dose-Volume Relationships for a Radiation therapy Treatment.

3099 Efficient Method to Train pDVH Models With Plan Quality Variation Present in the Training Cohort L. Appenzoller,1 J. Tan,1 D. Yang,1 P.W. Grigsby,1 J.K. Schwarz,1 S. Mutic,1 and K.L. Moore2; 1Washington University School of Medicine, Saint Louis, MO, 2University of California, San Diego, San Diego, MO Purpose/Objective(s): Plan quality variability is a known problem in clinical IMRT. The objective of this work was to develop a method to efficiently train accurate predictive DVH (pDVH) models for post-operative endometrial cancer patients when plan quality variations are present in the training cohort. Materials/Methods: A previously developed framework to predict achievable OAR DVHs that correlates expected doses to voxel distances from a PTV surface was used to create pDVH models for post-operative endometrial cancer patients. A random sample of 20 clinically treated IMRT plans with identical clinical objectives was used to train raw pDVH models for rectum, bladder, and sigmoid colon. A sum of residuals (SR) analysis quantifying the integrated difference between the clinical DVHs and the pDVHs showed larger plan quality variation than seen in previously modeled sites (prostate, head-and-neck). This cohort was used to test a method to train pDVH models that accurately predict OAR DVHs without replanning every sub-optimal patient. Initial training plans were ranked using the raw pDVH model’s mean SR for rectum, bladder, and sigmoid. The 5 worst ranked plans were replanned, improving OAR DVHs while maintaining PTV V95% > 100% and V115% < 0% per institutional standards. Replan_25% pDVH models were trained with the 5 best ranked plans and the 5 replanned outliers. The entire 20 patient cohort was replanned as a benchmark. These replans were also used to create