24. Does Mid-P CT image decrease inter-observer variability compared to MIP CT image?

Introduction As part of a phase 2 randomized clinical trial that compares Internal Target Volume (ITV) and Mid-Position (Mid-P) conformational planning strategies in locally advanced non-small cell lung cancer treatment, we investigated the impact of the type of CT image, i.e. Maximum Intensity Projection (MIP) image and Mid-P image on the inter-observer delineation variability. Methods We considered 12 patients among the 40 patients included in the clinical trial. For each patient, dataset for the delineation consisted in MIP image and Mid-P image, computed from four-dimensional CT scan (Big Bore, Philips). Primary tumor (IGTVT for MIP image and GTVT for Mid-P image) and pathological lymph nodes for 8 of the 12 patients (IGTVN for MIP image and GTVN for Mid-P image) were delineated on both images by an expert in lung radiation oncology and 2 or 3 radiation oncologists. The expert contours were considered as a reference and were compared to the others using paired Dice coefficient to estimate the overlap, Bidirectional Local Distance (BLD) to evaluate a Hausdorff distance-based between contours and the reference and Root-Mean-Square Deviation (RMSD) to measure volume variations with respect to the reference. Results The Mid-P delineated volumes were smaller than the MIP ones in average: 2.96 ± 2.75 cc and 16.7 ± 20.7 cc for nodes and primary tumors respectively. For the whole cohort, the DICE means and standard deviations were 0.685 ± 0.102, 0.706 ± 0.124, 0.777 ± 0.0926 and 0.784 ± 0.0756 for the GTVN, IGTVN, GTVT and IGTVT respectively. The absolute means BLD between the expert contours and the others were, in the same order, 2.08 ± 0.819 mm, 2.22 ± 1.36 mm, 2.60 ± 1.34 mm and 2.92 ± 1.35 mm. The RMSD were 3.05 ± 1.92 cc, 4.15 ± 4.66 cc, 13.4 ± 13.1 cc and 17.9 ± 18.3 cc respectively. Conclusions For one patient, we excluded an observer because of a non-complete contour. Regarding GTVN and IGTVN, we observed differences in terms of node selections and also delineation interpretations (nodes vs. lymph node areas). To ensure robust data, we only kept nodes selected by the expert. The presence of atelectasis and fibrosis led to large variations of primary tumor delineation (GTVT and IGTVT). The volumes and its variations decreased between MIP and Mid-P contours. Dice coefficients were lower with MIP compared to Mid-P. However, this metric is volume-dependent and tends to be lower for smaller volume. It may not be adapted here. The BLD was not statistically different. Finally, the volume variations were reduced but our results depended on delineation difficulties and did not demonstrate the statistical superiority of the Mid-P strategy in terms of inter-observer delineation variability.