Analyzing the impact of intrafraction motion: correlation of different dose metrics with changes in target D95%

A number of techniques are available to determine the dosimetric impact of intrafraction motion during intensity modulated radiation therapy (IMRT). Motion-induced dose perturbations can be determined both computationally and experimentally using a number of different dosimetric metrics. However, these measures may lead to different conclusions regarding the clinical impact of motion. This study compares the analysis of identical dose perturbations using different dosimetric metrics. Calculated changes in target D95% are used as a reference.

A total of 3768 motion-encoded dose distributions were calculated for nine lung tumor patients. The motion-encoded dose distributions were compared to static dose distributions using three dosimetric metrics: 2D gamma, 3D gamma, and histogram analysis. Each of these metrics was used to analyze dose perturbations both globally and within the target structure. Furthermore, the failing voxels were analyzed separately according to failure mode, i.e., under vs. over-dosed voxels. Metrics were evaluated based on their agreement with changes in target D95%. Evaluations included the metrics' maximum average sensitivity and specificity (MASS) in detecting unacceptable deliveries, a coefficient correlated to ranking (tau), and the linear correlation coefficient, r.

Of the evaluated metrics, the histogram metric restricted to the under-dosed voxels within the target agreed best with changes in target D95%. This metric achieved a MASS of 0.93, a tau of 0.69, and an r-value of 0.85. In comparison, the unrestricted 2D gamma metric achieved MASS = 0.77, tau = 0.40, and r = 0.67. Restricting the 2D gamma test both geographically and in failure mode increased the MASS to 0.85, tau to 0.70, and the r-value to 0.80.

This study suggests that any clinical decisions based solely on an unrestricted 2D gamma metric are suboptimal. A geographic and failure mode restriction can improve results. The remaining uncertainties with non-DVH (dose volume histogram) based metrics should be kept in mind when they are used to evaluate the dosimetric impact of target motion.