ESTRO 38 Abstract book

S1125 ESTRO 38

Figure 2 Conformity index for each post-training patient. Conclusion Radiologist training reduced delineation variability on all sequences and particularly on the MR sequences. Post- training the MEDIC sequence had the least delineation variability, was preferred by all delineators and produced slightly smaller volumes compared to CT. This study suggests the MEDIC sequence in conjunction with radiologist training can reduce the variability in prostate delineation. EP-2048 Development and implementation of a cost- effective technique to improve CT scans for contouring D. Nash 1 , A.T. Davis 1 , A.L. Palmer 1 1 Queen Alexandra Hospital, Medical Physics, Portsmouth, United Kingdom Purpose or Objective Radiotherapy CT scans are generally of inferior image quality (IQ) to a diagnostic CT scan. However, for IMRT and VMAT, more intricate structures need to be visualised, thus needing superior IQ. This is particularly true considering that the clinician outlining can be the most significant uncertainty in radiotherapy. This work was to develop a method of enhancing IQ for contouring without impairing the dose calculation, introducing any motion error or any additional imaging dose. Material and Methods The proposed method was to provide a second reconstruction of the planning CT raw data with alternative, optimised, parameters specifically to improve the image for contouring. The second scan was to be fused with the planning scan but never used for planning dose calculations. The preferred parameters were determined, with an initial focus on head and neck (H&N) scans. These were determined qualitatively by two experienced physicists by reconstructing previous scans acquired on a Canon Aquillion LB CT scanner, with a particular focus on edge and contrast enhancement. Reconstruction of the same data set meant no induced error due to motion between two scans, nor any additional radiation dose. Parameters changed included the reconstruction kernel to a head- specific one from the standard body one (FC13) and reduction of the reconstruction FOV diameter. The scan was then fused onto the planning CT scan and guidance sought from two experienced radiation oncologists. For the chosen parameters, the next 11 (6 with IV contrast) head and neck patients had this second scan reconstructed. Each oncologist was provided with a short questionnaire asking to qualitative opinions on the additional scan, whether there was any reduction in outlining time, their degree of confidence in images and how they compared to diagnostic scans stored for the patient on the PACS system. Comments were also requested. Results The selected image parameters for the secondary reconstruction were to reduce the FOV to approximately 20cm, from the usual 55 or 80 cm, and changing the reconstruction kernel to a sharper head kernel (FC44). An example is shown in fig. 1. For 9 of the 11 scans the Oncologists reported a reduction in outlining time, with no change for 2. For all scans the Oncologists had more confidence in their outlining. When compared against PACS, 9/11 were superior, 1/11 equivalent and 1 poorer (which was acquired at a different hospital). In terms of the comments given, all were positive. Based on these results the technique was adopted for all H&N patients and performed automatically with little workflow impact.

reviewed by an expert urology radiologist and discussed with the oncologists. The same oncologists subsequently delineated the second cohort. The delineation variability was quantified using a generalised conformity index for each image and patient. This is the intersection volume divided by the union volume generalised to be independent of the number of observers. The delineation variability for each MR sequence was compared to the CT within each cohort using a paired t-test. The mean ratio of volumes delineated on each MR sequence to the volumes delineated on CT was calculated. Results The radiologist training substantially improved the delineation variability for all three image data sets (see figure 1). Prior to radiologist training, the delineation on MEDIC sequence showed a similar variability to CT and the SPACE sequence significantly more (P = 0.05). Post- training delineation on the MEDIC sequence showed a significantly reduced variability compared to CT (P = 0.01) and the SPACE similar to CT. The MEDIC sequence gave the least delineation variability in 9/10 post-training patients (figure 2) and was preferred by all delineators. The volumes on both MR sequences were consistently smaller than on CT. Post-training the SPACE volume was 87 ± 2 % of the CT (mean ± s.e.m.). The MEDIC volume was 96 ± 3 % of the CT. Further work will investigate the reason for differences in volume size between the two MR sequences.

Figure 1 Mean conformity index pre and post radiologist training. 1 indicates perfect agreement. Error bars are one s.e.m.