B. Guo1, K. L. Stephans2, J. Z. Lee3, and P. Xia4; 1Department of Radiation Oncology, Cleveland Clinic Foundation, Cleveland, OH, 2Department of Radiation Oncology, Cleveland Clinic, Cleveland, OH, 3Department of Cardiovascular Medicine, Cleveland Clinic, Cleveland, OH, 4Department of Radiation Oncology, Taussig Cancer Institute, Cleveland Clinic, Cleveland, OH
Purpose/Objective(s): Cardiac radioablation is an emerging therapy for recurrent ventricular tachycardia. Target delineation on simulation CTs for cardiac radioablation is challenging and error-prone. This study reports the results of an in-house script developed to (1) assist angulation of the left ventricle (LV) to standard cardiac view, and (2) automatically create the seventeen segments of the LV defined by American Heart Association. Materials/
Methods: The semi-automatic script uses the LV contour as the initial input, guides the user to rotate the simulation CT to standard cardiac view, and then automatically creates the 17 segments of the LV in two steps. First, ventricle wall structure is created based on the Hounsfield Unit (HU) differences of the LV wall and cavity. Then a template is loaded to divide the wall into 17 segment volumes defined by the American Heart Association. A cardiologist reviews and edits the target contours of the segments intended to be treated in the cardiac view. Eight patients were included in this retrospective analysis. On average 6.3 (range 2-10) of the 17 segments were treated per patient. Six patients were simulated free breathing with 4DCT and the other two were simulated with deep inspiration breath hold. IV contrast was injected at 1 ml/s during simulation scans. The auto-delineated target volumes by the script were compared with the final target volumes approved by the cardiologist using dice similarity coefficient. Results: The semi-automatic script took less than 5 minutes to complete. For the 8 patients, angulation rotation was -560 +/-110 in the axial plane, -670 +/-100 in the sagittal plane, and -40 +/- 100 in the coronal plane. The mean ventricle wall thickness was 1.0 +/- 0.1 cm. The dice coefficient between the auto-segmented target volumes and the final target contours was 0.60 +/- 0.23. Dice coefficient had a moderate correlation (R=0.55) with the mean HU difference between the LV wall and cavity, suggesting that the contrast of the simulation CT affects the accuracy of the automatic LV wall and segment volume creation. Conclusion: A semi-automatic script was developed to assist angulation of the simulation CT images to standard cardiac view and to automatically create the 17 segments of the left ventricle. The script can improve the efficiency and consistency of target delineation for cardiac radioablation.
