Does the registration of FDG-PET and planning CT images decrease inter- and intra-observer variation in delineating tumor volumes for non-small cell lung cancer (NSCLC)

Abstract

Multiple studies have shown that there is an advantage to incorporating FDG-PET imaging studies in the determination of tumor volumes for radiation therapy (RT). This study investigates whether further benefit is gained by registering the planning CT with the FDG-PET study when contouring tumor volumes as compared with using side-by-side image sets. Nineteen patients with NSCLC had FDG-PET scans in the treatment position which were then registered with treatment planning CT images. Stage of disease was I/II–26%, IIIA–42%, IIIB–32%. Two radiation oncologists who had not previously treated these patients contoured tumor volumes for 9 patients using registered FDG-PET and CT images (“registered”) and for 10 patients using separate FDG-PET images as a guide (“non-registered”). Each pair of volumes was compared both quantitatively and qualitatively. For quantitative comparison, the treatment planning computer calculated the concordance index, defined as the ratio of the intersection (A and B) of the 2 volumes to their union (A or B); ie, concordance = (A and B)/(A or B). For qualitative analysis, the planning system constructed an AP and lateral digitally reconstructed radiograph (DRR) for each patient and projected the pairs of volumes onto it. Differences in volumes were graded as insignificant, minor, moderate or major based on variations in the tumor volume in all three dimensions, taking into account the clinical significance of the areas in question. Any inclusion of a lymph node region on one contour but not the other was considered a major difference. To assess intra-observer variation, a third radiation oncologist contoured 2 sets of tumor volumes for all patients. This investigator had previously treated these patients, and there was a median of 40 months between the time the first and second tumor volumes were delineated. Ten of the new contours were repeated using these same registered images (“registered/registered”) and 9 were re-done without registration (“registered/non-registered”). Each pair of tumor volumes was evaluated quantitatively and qualitatively as described above. Between the 2 radiation oncologists, the median inter-observer percent concordance among pairs of tumor volumes in the non-registered group was 61%, and was 70% in the registered group (p = 0.001). On qualitative analysis, of the pairs of volumes in the non-registered group, 5/10 differences were insignificant, 3/10 were minor and 2/10 were moderate. The differences within pairs of volumes in the registered group were mostly insignificant (7/9), with the remaining 2/9 pairs having only minor differences. For the physician who contoured each patient twice, the median intra-observer percent concordance among pairs of volumes in the registered/non-registered group was 58%, and among those in the registered/registered group it was 71% (p < 0.05). On qualitative analysis, intra-observer differences in volumes in the registered/non-registered group were graded as insignificant in 2/9, minor in 2/9, moderate in 0/9 and major in 5/9. Among those pairs in the registered/registered group, differences were graded as insignificant in 2/10, minor in 6/10, moderate in 2/10 and major in 0/10. The registration of FDG-PET and planning CT images for target definition in NSCLC decreased both the inter- and intra-observer variability in the determination of tumor volumes. This implies that greater accuracy in tumor delineation can be achieved by registering FDG-PET and CT images for contours rather than CT images with separate FDG-PET images as a guide