ESTRO 38 Abstract book

S151 ESTRO 38

intensity variation correlation with a set of neighboring pixels. Next, an enhanced heart edge image is generated by multiplying the 1-3 Hz intensity variation of each pixel with the correlation of intensity variations between the pixel and the heart edge candidate pixel (Figure 1B). Finally, the heart edge is segmented in the enhanced heart edge image and the exposed heart area is calculated as the area between the heart edge and the medial field edge (Figure 1C).

tracking clinical trial (NCT02514512). Each patient had three electromagnetic beacons (Calypso) inserted into the lung surrounding the tumor. An MLC tracking SABR plan was generated with the planning target volume (PTV) expanded 5mm from the end-exhale gross tumor volume (GTV). For comparison a conventional motion- encompassing SABR plan was generated with PTV expanded 5mm from a 4DCT-derived internal target volume (ITV). Treatment was delivered using a standard linear accelerator using in-house developed software to continuously adapt the MLC motion based on the Calypso beacons’ movement. The rate of successful treatment fractions with MLC tracking, tumor motion, treated volume and reconstructed delivered dose were compared between MLC tracking and conventional ITV treatment planning. Results All seventeen patients were treated successfully with MLC tracking (70 successful fractions), completing the primary endpoint of this study. Tumor motion range varied during treatment, between fractions and from the planning 4DCT; significantly, larger motion was observed during treatment that exceeded standard PTV boundaries. The MLC tracking PTVs for all patients was smaller than with ITV based planning (mean 29%, range 2%-47% reduction, or 2-18 cm 3 with MLC tracking). Subsequent reductions in normal lung dose were observed. Reconstruction of delivered treatments confirmed accurate delivery of MLC tracking, with 100% of the prescribed dose delivered to the GTV for all 70 fractions. Conclusion Real-time adaptation via MLC tracking on a standard linac has been successfully performed in seventeen lung cancer patients. Reductions in treated volumes up to 47% were observed, which translated to reductions in lung dose. OC-0299 Fully automatic detection of heart irradiation in cine MV images during breast cancer radiotherapy P.R. Poulsen 1 , M.S. Thomsen 2 , R. Hansen 2 , E. Worm 2 , E. Yates 2 , H. Spejlborg 2 , B. Offersen 1 1 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark; 2 Aarhus University Hospital, Department of Medical Physics, Aarhus, Denmark Purpose or Objective Heart irradiation during radiotherapy of breast cancer can lead to late cardiac morbidity and increased mortality for long-time survivors. Left-sided breast cancer patients are therefore often treated in deep-inspiration breath-hold (DIBH) to better separate the breast and the heart. However, daily variations in the heart position can give heart exposures that are much larger than planned. For tangential treatments, continuous portal images (cine MV images) may be used to monitor the heart exposure at each treatment fraction with no additional dose to the patient. Challenges include low image contrast and the location of the heart edge in or near the field penumbra. In this study, we develop and test automated heart detection in cine MV images. Material and Methods Cine MV portal images (Figure 1A) of 302 tangential field deliveries were recorded at 7.7 Hz for ten left-sided breast cancer patients who received DIBH radiotherapy in 15-18 fractions. An algorithm for fully automatic detection of the heart edge in cine MV images was developed and tested for all available images. The algorithm exploits that the intensity of pixels at the edge of the heart will change cyclically with frequencies of 1-3 Hz due to heartbeat and that the intensity changes of all pixels at the heart edge will be highly correlated with each other because they have the same physical origin (heartbeat). The algorithm first identifies a candidate pixel on the heart edge as the pixel in the image with highest value of (1) intensity variations in the 1-3 Hz frequency range multiplied by (2)

Results Part of the heart was exposed at 169 out of 302 field deliveries. Using all cine MV images, the heart edge was correctly identified in all cine MV series except for 11 cases with heart drift motion during field delivery. For these cases, analysis of a shorter time series with less drift motion (30-50% of the MV images) correctly identified the heart edge. Figure 2 shows an example of the exposed heart area for a patient. While large interfraction variations occurred the intrafraction variations were smaller with high correlation between the heart exposure at field 1 and field 2 at a fraction (r = 0.85, p < 0.001).

Conclusion An algorithm for automatic identification of pixels at the heart edge in cine MV images was proposed, developed and shown to be highly efficient for heart exposure detection in tangential breast fields. The algorithm can be