DOS 543 - Discussion 2
Post: Biologically Based Treatment Planning Systems
The AAPM TG-166 report on biologically based treatment planning systems (BBTPS) discusses the use of tumor control probability (TCP) and normal tissue complication probability (NTCP) models in the design of treatment plans.1 The simplest models used for treatment design are based on correlations of outcomes to physical dose levels. More nuanced approaches such as TCP and NTCP try to incorporate data about the radiobiological sensitity of different types of tissue and different types of tumors. Generally speaking, more radiation is associated with more tumor control, but it is also associated with more tissue damage. The goal of a BBTPS is to optimize a plan such that it finds a balance between high confidence of tumor control and sparing of critical structures.
Some of the challenges of designing a BBTPS revolve around balancing the importance of multiple independent constraints. Algorithms that seek improvement by making successive small changes often find a locally optimized solution that is quite different from the global best possible solution. A robot trained to always walk uphill will always find the top of the current section of terrain, but it may only crest a small foothill when there is a much taller mountain nearby. Whether or not the robot finds the top of the large mountain depends on where it starts its search and what algorithm it uses to traverse the terrain. In radiation planning terms, the multiple tumor constraints and normal tissue constraints are the factors that shape the terrain up and down.
Another complicating factor is that models that optimize for one dimension often create problems that must then be addressed with another model. For instance, models that use TCP have no reason to seek dose uniformity, because more dose always correlates with more control. In order to compensate for this, other constraints, such as maximum physical dose, must be applied. When optimizing dose coverage across organs at risk, NTCP models can take into account the functional subunit arrangement of each organ. Organs arranged in serial fashion, where damage to a single functional subunit can destroy the function of the entire organ, need to be optimized quite differently from organs arranged in parallel fashion, where the dominant goal is to reduce the number of subunits affected. Dose optimizations on serial organs will place a heavy emphasis on not exceeding a pre-determined maxmium dose level. This reduces the likelihood that any subunits will be compromised. The tradeoff is that such models do not simultaneously control the number of subunits being exposed and they do not care as much about low and medium range doses. This may result in a solution that delivers more low-dose than was necessary. Optimizers that protect parallel organs behave the other way. They optimize in such a way as to maximize the number of subunits that are preserved. They do this by applying a strong penalty function at medium dose levels, but allowing the penalty function to fall to zero once medium dose levels have been exceeded. The logic is that any subunits that are already compromised can not be further hurt by more radiation, so it is better to let excess dose traverse those areas instead of fresh tissue.
As a dosimetrist, I use several of these concepts already, but in a manual fashion. When I am evaluating a plan that passes dose through the spinal cord (a serial organ), I always focus on the maximum dose, but I do not always pay close attention to the lower dose ranges. When I am building plans that pass beams through the lungs, I pay attention to the volume of lung receiving 20 Gy, 10 Gy, and 5 Gy, but I do not concern myself with the volume receiving 40+ Gy. Upon reading the TG-166 report, I am now seeing that I should be paying closer attention to both dose ranges on every organ. This is not always practical, since humans can only concentrate on so many interrelated factors at a time. A computer can perform these complex balancing acts with ease, but only if the desired conditions are correctly modeled and optimized. That is the reason that BBTPS systems were created. The planning system I use during my internship, Varian Eclipse, has several models available, but I have never explored the options available to me. With this report in hand, it will be possible to explore those options and better understand what they mean.
- Li X, Alber M, Deasy J, et al. The use and QA of biologically related models for treatment planning: short report of the TG-166 of the therapy physics committee of the AAPM. Med Phys: 2012;29(3):1386-1409. http://dx.doi.org/10.1118/1.3685447.