Fine-tuning iPDT DosimetryIntroduction of Clinical Considerations to the Process of iPDT Treatment Planning in PDT-SPACE March 2023
Shuran WANG 1, Tina SAEIDI 2, Lothar LILGE 2, Vaughn BETZ 1
ABSTRACT
The use of interstitial Photodynamic Therapy (iPDT) in the treatment of deeply-seated tumors has shown promising results as a minimally invasive oncology treatment. PDT-SPACE is an open-source software tool that automates the process of personalized iPDT treatment planning by providing a placement of light probes optimized for the provided tumor shape and location to minimize damage to the surrounding healthy organs at risk (OAR). However, many placements are not realizable in clinical settings as the optimized placement often requires the insertion of light probes in sensitive, or multiple directions. Furthermore, the optimization result from PDT-SPACE relies heavily on a good initial probe placement provided by the user, posing more difficulties in clinical usage. Here we propose to improve PDT-SPACE by allowing more user control through the use of clinical constraints, as well as the addition of an algorithm that generates reasonable initial probe placements as the starting point for PDT-SPACE's refinement algorithms to further automate the process.
To specify the clinical constraints, we introduced a new XML file format, which allows clinicians to specify options such as the initial probe placement, maximum allowable length for each probe, the point of injection for all probes, and whether each option is enabled. The XML file format is self-documenting and flexible to allow additional options in the future. The main clinical constraint added here is the constraint for the injection point, specified by a circle with a normal vector pointing out of the surface and a diameter; this defines a burr hole on the skull surface for brain tumor indications or an entry area for other indications. The software will be constrained to generating probe placements passing through the circle; thus only a single burr hole is needed for the operation.
The automatic generation of an initial probe placement is introduced in two modes depending on whether an injection constraint is specified. In the case where the injection point is constrained, the heuristic is to place the first probe along the line extending from the injection circle center to the tumor center, and then define a grid of parallel probes passing through both the injection circle and the tumor. On the other hand, if the injection point is not constrained, the top priority is to minimize the number of probes used, so the generated initial placement will orient along the longest axis of the tumor shape. The longest axis is approximated by computing the oriented bounding box using the Principle Component Analysis algorithm; the probes will be placed along the direction of the longest boundary. In either case, PDT-SPACE's iterative refinement algorithms then further optimize the probe placement. Results will be collected by simulating 9 brain tumor models created from the Colin27 brain mesh; the simulations run 10 times for each model with different random seed starting points to account for variability in the optimization algorithms.