Comparison of intensity-modulated radiotherapy planning based on manual and automatically generated contours using deformable image registration in four-dimensional computed tomography of lung cancer patients.

PURPOSE: To evaluate the implications of differences between contours drawn manually and contours generated automatically by deformable image registration for four-dimensional (4D) treatment planning. METHODS AND MATERIALS: In 12 lung cancer patients intensity-modulated radiotherapy (IMRT) planning was performed for both manual contours and automatically generated ("auto") contours in mid and peak expiration of 4D computed tomography scans, with the manual contours in peak inspiration serving as the reference for the displacement vector fields. Manual and auto plans were analyzed with respect to their coverage of the manual contours, which were assumed to represent the anatomically correct volumes.
RESULTS: Auto contours were on average larger than manual contours by up to 9%. Objective scores, D(2%) and D(98%) of the planning target volume, homogeneity and conformity indices, and coverage of normal tissue structures (lungs, heart, esophagus, spinal cord) at defined dose levels were not significantly different between plans (p = 0.22-0.94). Differences were statistically insignificant for the generalized equivalent uniform dose of the planning target volume (p = 0.19-0.94) and normal tissue complication probabilities for lung and esophagus (p = 0.13-0.47). Dosimetric differences >2% or >1 Gy were more frequent in patients with auto/manual volume differences > or =10% (p = 0.04).
CONCLUSIONS: The applied deformable image registration algorithm produces clinically plausible auto contours in the majority of structures. At this stage clinical supervision of the auto contouring process is required, and manual interventions may become necessary. Before routine use, further investigations are required, particularly to reduce imaging artifacts.