Gamma Knife (GK) radiosurgery treats brain lesions through multiple targeted
radiation exposures of varying duration and spatial distribution. Clinical
radiosurgery treatment planning only considers the total amount of delivered
radiation. A biologically effective dose (BED) model allows quantifying the
damage induced in a tissue due to radiation exposure while accounting for
cellular repair. With this thesis work, we explore the potential and feasibility
of using the more complex BED formulation to generate biologically-aware
treatment plans.
To this end, we quantify the impact of changes in the temporal domain
of treatment delivery (i.e. beam-off periods, order of delivery), which need
to be considered at the treatment planning stage to avoid undesirable BED
variations. The delivery sequence alone can incur variations in marginal BED
by up to 14%.
Consideration of treatment delivery timing and sequence creates a nonconvex nonlinear treatment planning problem that is too computationally
expensive to solve in a time-sensitive clinical setting. We develop multiple
optimisation techniques to identify the most suitable one for a clinical workflow.
While a convex underestimator approach provides slightly improved solutions,
it requires several orders of magnitude more computational resources than
local optimisation approaches that reach similar performance in terms of plan
quality.
In consultation with our clinical collaborators, we devise a BED treatment
planning workflow that further reduces the possible planning times by combining pre-computation of candidate solutions with interactive exploration and
refinement of the final treatment plans. To evaluate this workflow, we develop
a prototype treatment planning framework. We show that BED optimisation
removes the time dependence and further increases plan quality. The results of
the proof-of-concept workflow demonstrate the feasibility of a future clinical
application of BED planning in GK radiosurgery.