A conceptual model integrating spatial information to assess target volume coverage for IMRT treatment planning

We propose a model that integrates the spatial location of each voxel within a clinical target volume (CTV) to differentiate the merit of intensity-modulated radiation therapy (IMRT) plans with similar dose-volume histogram (DVH). This conceptual model is based on the hypothesis that various subregions within a given CTV that may carry different degree of risk in containing microscopic disease.

We hypothesize that a correlation between the probability of microscopic tumor extension and the risk of lymph node metastasis of a particular voxel point within CTV can be inferred based on its distance from the surface of radiographically evident gross disease. A preliminary observation was gathered from existing clinicopathologic data, and, based on these observations, a conceptual model for exponential-decay microscopic-extension probability function around primary tumor and linear function parameters relating the likelihood of lymph node metastasis to the distance from primary tumor was proposed. This model was generated to provide scoring functions to examine the merit of IMRT plans. To test the feasibility of this model, we generated two IMRT plans with similar and clinically acceptable DVH-based CTV coverage. Planning data were transferred to a data analysis software package (Matlab, The Mathworks Inc.). A 3D scoring function was calculated for each voxel inside the CTV. The adequacy of target coverage was evaluated by several novel approaches: 2D dose-volume scoring-function histograms (DVSH), the integral probability of relative residual tumor burden (RRTB), and tumor control probabilities (TCP) employing the scoring function as a pseudo-clonogen density distribution.

Incorporating parameters for the risk of containing microscopic disease in each voxel into the scoring function algorithm, 2D DVSHs, RRTBs, integral RRTBs, and TCPs were computed. On each axial image, an RRTB map could locate the regions at greatest risk. These scoring functions were able to differentiate the merit of CTV coverage of clinically different IMRT plans but having very similar DVHs; one with cold spots centrally located over the gross tumor, and the other with more acceptable cold spots on the periphery of the CTV further away from the gross tumor volume.

We demonstrated the feasibility and potential utility of an IMRT scoring method derived from this conceptual modeling approach. These methods are capable of ranking treatment plans with similar DVH profiles but different underdosed regions within the target. We will examine the accuracy of model parameters by performing tumor-specific image-pathologic correlation studies. Upon validation of these parameters, incorporating this scoring function model into plan optimization may have the potential to avoid underdosing subvolumes within CTV that harbor a higher likelihood of microscopic disease.