Image-based modeling of normal tissue complication probability for radiation therapy.

We therefore conclude that NTCP models, at least in some cases, are definitely tools not toys. However, like any good tool, they can be abused and in fact could lead to injury with misuse. In particular, we have pointed out that it is risky indeed to apply NTCP models to dose distributions which are very dissimilar to the dose distributions for which the NTCP model has been validated. While this warning is somewhat fuzzy, it is clear that more research needs to be done in this area. We believe that ultimately for NTCP models to be used routinely in treatment planning in a safe and effective way, the actual application will need to be closely related to the characteristics of the data sets and the uncertainties of the treatment parameters in the models under consideration. Another sign that NTCP models are becoming tools rather than toys is that there is often good agreement as to what constitutes a correct direction of improving the reduced risk for that particular complication endpoint. Thus, for example, mean dose to normal lung almost always comes out as being the most predictive or nearly most predictive factor in the analysis of radiation pneumonitis.