ESTRO 38 Abstract book

S1119 ESTRO 38

The above results show that there is significant inter‐ observer variation in outlining of the rectum on MVCT images. In the context of this variation, DIR appears to produce outlines that are comparable to those of oncologists. These outlines may also be a superior alternative to the automatic segmentation algorithm currently used for this purpose. EP-2039 Validation Of The Direct Density™ Ct Image Reconstruction Algorithm I. Peiro Riera 1 , E. Fernandez‐Velilla Ceprià 1 , J. Quera Jordana 1 , O. Pera Cegarra 1 , N. Anton Comelles 1 , M. Prieto Carballo 1 , M. Algara López 1 1 Hospital del Mar, Radiotherapy Oncology, Barcelona, Spain Purpose or Objective The standard kV setting for Simulation CT images in radiotherapy is 120kV. Planning Systems (TPS) use calibration curves at that energy for associating electron densities to Hounsfield units (HU). For contrast and radiation protection reasons, in some situations a different kV may be recommendable. This requires the commissioning of one calibration curve for each kV. In our work we analyze the Siemens reconstruction algorithm Direct Density™ (DD), which gives CT numbers proportional to relative electron densities (RED), independent of the kilo‐voltage used. DD is evaluated and compared to the standard Filtered Back‐Projection (FBP) in order to verify its accuracy to be introduced in the radiotherapy workflow. Material and Methods A Siemens Somatom Confidence RT Plus CT system was used to acquire the images at 70, 80, 100, 120 and 140kV. Slice thickness and pitch were the same used in clinical protocols for pelvis and brain patients. mAs were adjusted at all kV in order to keep CTDI unchanged from the reference 120kV studies. A Catphan CTP 604 phantom was scanned in order to evaluate image quality in terms of high and low contrast resolutions, uniformity and noise. Additionally, a Gammex 467 Tissue Characterization Phantom was scanned to obtain calibration curves and to confirm that they were kV independent when DD was applied. CTDI doses were verified with an RTI Piranha to assure they were unchanged. Finally, a comparison between 3D doses using FBP at 120kV and DD images was done by means of a Varian Eclipse v. 13.6 TPS in real patients, using both AAA and Acuros XB 13.7 algorithms. Results The observed differences in calibration curves were maximal at 70kV for the cortical bone and the 50% CaCO 3 rods (highest densities). For the other values of kV, HU and electronic density showed the same linear relationship whatever the kV was. Image quality was slightly better in FBP than in DD images (Table 1). Spatial resolution in high contrast showed a worsening from 0.71mm to 0.83mm, and in low contrast no circles of the Catphan module could be detected in DD images at energies over 70kV. A slight difference in uniformity was observed, always being FBP better than DD. Noise was significantly higher in the DD images, around 3 times higher. As expected, DD does not result in significant differences in dose calculations in comparison with FBP. Using AAA, the maximum differences were 0.1% in pelvis and 0.7% in brain. For Acuros the same differences were found.

Purpose or Objective We have evaluated deformable image registration (DIR) as a method of automatically outlining the rectum on pelvic megavoltage CT (MVCT) scans. The rationale for this work is to retrospectively correlate the dose delivered to the rectum with the toxicity patients experience from external beam radiotherapy for prostate cancer. To calculate the dose delivered to the rectum, it must first be delineated on MVCT images acquired at the time of the patient’s treatment. However, the study recruited 529 prostate patients who each received 20 or 37 treatment fractions; manual outlining for such a large number of images would clearly be impractical. Our group currently uses a purpose‐built automatic segmentation algorithm to outline the rectum on these images. However, DIR presents a potential alternative method of doing this, and we wished to investigate its performance for this purpose. Material and Methods On six MVCT images from six different patients, the rectum was independently outlined by ten oncologists, according to our departmental protocols. They were also outlined using the automatic segmentation algorithm. The rectums on patients’ original planning CT scans had been outlined by one oncologist at the time of treatment planning. Three pieces of DIR‐capable software were tested as part of this project: ‐ Prosoma (MedCom, Darmstadt, Germany) ‐ Pinnacle (Philips, Amsterdam, The Netherlands) ‐ RTx (Mirada, Oxford, United Kingdom) Each piece of software was used to deformably register the patients’ planning CT scans to their MVCT scans. The calculated deformations were then applied to the original rectum outline on the planning CT to propagate it to the MVCT dataset. Contours were compared to one another using the Jaccard Index and Hausdorff Distance, to assess general volume overlap and gross spatial variation. Each oncologist’s outline was compared to all others for each patient, to assess inter‐observer variation in outlining. The outlines produced using DIR and automatic segmentation were then compared to each of the oncologists’ outlines to assess the performance of the software. Results The mean Jaccard Indices (JI) and Hausdorff Distances (HD) between the oncologists’ outlines, and between outlines from each piece of software and each oncologist, are displayed in Table 1.

The figure below displays the variation in HD between each pair of oncologists for each of the six patients in this study. The central bar indicates the mean HD for that patient, while the smaller bars indicate the range of the mean ± one standard deviation.

Conclusion