The use of small planning-target-volume (PTV) margins and highly conformal dose distributions decreases the likelihood of any sub-clinical disease not included in the clinical target volume (CTV) being coincidentally controlled. And where advanced imaging modalities become more and more accurate in the localization and characterization of the primary tumour, detection of microscopic disease remains notoriously problematic. Simply increasing the size of the CTV to compensate for this risk would effectively close down the clinical window again - having painstakingly opened it up by means of advanced imaging and treatment-delivery techniques.
It's commonly assumed that the small number of undetectable clonogenic cells don't require the same high dose as the primary tumour to be eradicated, so increasing the size of the CTV would probably be overkill and introduce toxicity that could have been avoided. A common scheme involves multiple CTVs with different margins and dose prescriptions, but this complicates treatment planning, and requires ad-hoc decisions on the sizes of these margins and the levels of dose prescriptions. One alternative solution that we are exploring is to replace the concept of the CTV as a fixed volume with a more flexible description of the target.
Biological control
Various groups have already addressed the idea of robust IMRT optimization, in which knowledge about the distributions of systematic and random errors is used directly by the objective functions that steer the dose optimization. This way, the use of PTV margins is no longer necessary, resulting in a more flexible treatment-planning process in which patient-specific knowledge of geometric variations can more easily be exploited.
The experimental implementation of robust optimization developed at the Netherlands Cancer Institute (NKI) places the biological TCP (tumour control probability) and NTCP (normal-tissue complication probability) models in a central role. This leads to a planning system that provides the best expected tumour control for a given expected rate of complications. From point to point in the patient's anatomy, the system weighs the risk of inducing toxicity against the risk of missing or underdosing the target.
The result is a dose distribution that encompasses the target with a large implicit margin where possible, but becomes tighter where nearby organs are at risk (figure 1). In the mechanistic TCP model used at the NKI, the target is represented by a distribution of clonogenic cells, and this distribution need not be uniform within a given volume. Furthermore, it's possible to work with probabilities of clonogenic cells being present at a given location. In other words, the description of the target can be probabilistic in nature, quite unlike the fixed concept of a CTV.
Data input
This planning technique may offer a flexible platform for the incorporation of patient-specific biological and geometrical information, but the models that it relies upon are not yet well enough developed for clinical release. To be able to strike a balance between expected treatment outcome and adverse treatment effects, these models should be provided with solid parameters that describe the processes of killing cells and damaging tissues, as well as an indication of where to expect cells that are not visible during imaging.
One place to look for such indications is in historical patient data. Recent analysis of archived treatment-planning data and clinical-outcome statistics for patients in a large randomized prostate trial indicated that high-risk patients without recurrence had received, on average, significantly higher dose to areas in the obturatorial and pre-sacral regions than patients who suffered a biochemical failure. Some patients happened to have received a somewhat higher dose to these areas, which are situated a few centimetres away from the prostate (figure 2). Such dose variations were not intended, but rather emerged as a side effect of the treatment techniques that were used.
Presumably, the higher dose to these areas was of benefit to these patients because there were cells there to be killed. While the interpretation of such dose differences comes with many uncertainties, analyses like these provide valuable information on the whereabouts of the clonogens we can't see, and on the dose necessary to eradicate them.
Eventually, work like this should result in the development of an advanced patient model, in which the locations and numbers of clonogenic cells, and their radiosensitivities are expressed, together with the uncertainties in these properties. Existing and future (biological) imaging modalities should then be used to narrow down these quantities for each individual patient, and decrease the associated uncertainties where possible.
As the treatment-planning process can be based upon this model, and can take its associated uncertainties explicitly into account, the resulting plan will constitute the most reasonable treatment-planning compromise, given the available data. As experience grows and more patient data and outcome statistics become available, the model can be updated and the associated uncertainties decrease. The resulting treatment-planning compromise will then approach the optimal plan for each patient ever more closely.