Session Item

A detailed contouring atlas of the heart valves is lacking. Existing heart contouring atlases have not been evaluated on average intensity projection four-dimensional noncontrast computed tomography (AVE 4D-CT) scans, routinely used for organ-at-risk delineation in lung cancer radiotherapy (RT). As a first step of a planned prospective trial on imaging-based evaluation of RT-related cardiotoxicity in NSCLC, we aimed to develop the heart valves contouring atlas and to assess interobserver variation in delineation of the heart, its substructures, and coronary arteries on AVE 4D-CT scans, along with the impact of contour variation on estimated RT doses.

A draft of the heart valves contouring atlas with written guidelines was developed by a radiation oncologist (JS) and agreed by cardiologists. Five radiation oncologists and four cardiologists were recruited to delineate the valves according to the draft, and the remaining heart substructures (4 chambers, 4 left ventricle segments, 4 coronary arteries and the heart) based on the existing heart contouring atlases on AVE 4D-CT scans of ten patients who underwent radio(chemo)therapy for NSCLC. Based on the delineation exercise, some modifications of the initial draft of the heart valves atlas were made and their delineation was repeated. Next, the "reference" contours for each structure and case were collectively defined, to which all the observer contours were compared. Spatial variation was assessed using the Sørensen-Dice similarity coefficient (DSC) and the directed average Hausdorff distance (DAH). The effect of spatial variation on RT doses was assessed using the patients' RT treatment plans. The results presented below are for the patient with the worst reproducibility.

A detailed contouring atlas of the heart valves was developed (Fig.1) and evaluated on AVE 4D-CT scans: inter-observer contour overlap (mean DSC) was 0.62, 0.55, 0.43 and 0.36, and interobserver contour separation (mean DAH) was 2.3, 3.4, 2.7 and 3.6 mm for the pulmonic, aortic, mitral and tricuspid valve, respectively. For the heart and its remaining substructures, defined according to the existing atlases, the interobserver agreement was the highest for the heart and its four chambers, lower for the left ventricular segments and the lowest for the coronary arteries (Tab.1). In the presented patient, this spatial variation resulted in <1.5Gy dose variation for 11 of 17 heart substructures contoured by nine observers on one AVE 4D-CT dataset; the highest variation resulting in the difference of 7.64Gy between the minimum and maximum estimated dose was recorded for the left main coronary artery.

Our atlas enables reproducible delineation of the heart valves. Delineation of the heart and its substructures on AVE 4D-CT scans according to the existing atlases is feasible, with inter-observer variability similar to that reported in validation studies of these atlases on conventional noncontrast CT scans.