Millimetric control of the penetration depth of protons within a tumor of a patient at which proton beams deposit maximum dose is key to the success of a treatment of cancer by proton therapy. This is achieved by modulation of their energy but highly vulnerable to uncertainties and anatomical changes. Generous safety margins around the target volume are currently used to consider these uncertainties, causing substantial dose deposition to normal tissue. Therefore, an in-vivo verification of the proton range in the patient is highly desirable. A promising approach is prompt γ-ray imaging (PGI). This technique was for the first time clinically implemented in Dresden in 2015 using a so-called slit-camera prototype developed in the EU project ENVISION under the lead of IBA. Today, PGI is routinely applied in Dresden within a clinical study (PRIMA) with the goal to systematically determine its clinical benefit. In-room control CTs acquired in treatment position serve as ground truth information for potentially occurring anatomical changes. For each treatment acquisition, the PGI system delivers data that is rich in conclusions that can prove complex to disentangle and thereby suited to AI-based classification models. For online-adaptive proton therapy, a close-to-real time automated processing and interpretation of the prompt-gamma-based treatment verification data is crucial and one of the main goals of the project.

The goal of the project is to develop a near-real-time automated decision-making tool based on prompt gamma information on clinically relevant treatment deviation and their source. This will ultimately allow a reduction of currently used safety margins, reducing the dose delivered to normal tissues, thus potentially reducing side-effects.

In the first part of the project the student will participate in the data acquisition and evaluation of PGI within the PRIMA study, aiming at automatize the PGI data processing workflow. This will fundamentally pave the way for its usability as near real-time treatment verification. Next, the real-time capability of the approach will be assessed, by evaluating (1) the processing time of the PGI and related treatment data, (2) the time for the spot-wise range shift determination and (3) the time for the automated AI-based decision making. In addition, the potential margin reduction of PGI treatment verification will be quantified and an automated confirmation of a PGI-triggered treatment intervention due to a clinically relevant deviation will be developed. Finally, the PGI-based range verification will be compared with the alternative range probing approach, developed by UMCG, on phantom level as enabled by RAPTOR synergy, potentially using the end-to-end phantom developed at MedUniVienna.