• hadrontherapy
• particle-therapy
• inside-project
• pet
• in-beam pet
• pet-monitoring
• p-value

In particle therapy, beams of energetic protons or charged ions are used for cancer treatment. The main advantage of using heavy charged particles, rather than conventional radiotherapy, is the convenient dose deposition. Heavy charged particles, indeed, release the majority of their energy in a very well defined position at a certain depth, called the Bragg Peak. This allows to deliver a high dose to tumors located at a certain depth inside the body, while sparing the surrounding healthy tissues.

Due to the highly selective energy deposition in particle therapy, a very precise dose monitoring system is of great importance to prevent under-dosage to the tumor and over-dosage to organs at risk (OARs), that can occur as a result of anatomical variations and other uncertainties in dose delivery. This is particularly important for patients affected by tumors in the head and neck region, that commonly experience anatomical variations during the treatment. For these patients, various OARs, such as spinal cord or salivary glands, are typically located near the treated regions, and over-dosage can lead to significant side effects.

One of the available methods for dose monitoring in particle therapy is Positron Emission Tomography (PET) which is based on the detection of the back-to-back photon pairs generated in the interaction between the tissue atomic electrons and the positrons emitted by positron emitting isotopes like carbon-11, oxygen-15 or nitrogen-13. The PET monitoring performed in real-time during the treatments is called in-beam PET.

The INSIDE (INnovative Solutions for In-beam DosimEtry in hadrontherapy) project realized an in-beam planar PET system, aimed to provide feedback on the dose deposition, together with a Monte Carlo (MC) simulation tool and an image reconstruction tool to predict the corresponding reference PET images. The system is currently in a clinical trial at the National Centre of Oncological Hadrontherapy (CNAO) center in Pavia.

In this master thesis we propose a new real-time method to process PET monitoring data, that allows to highlight possible morphological changes in the patient body occurring in between the treatment sessions. The method was developed and tested on MC simulations to address its feasibility and its expected sensitivity.

First, a patient treated at CNAO for a Squamous Cell Carcinoma at the left sinonasal cavity was selected. For that patient, the control CT acquired after 8 weeks from the beginning of the treatment, highlighted a significant morphological change that determined the need of a replanning of the treatment.

Second, the level of stochastic fluctuations in the PET image was analysed for each voxel, by simulating a large number (102) of PET images, based on the Computed Tomography (CT) exam of the patient taken before treatment. In this way we obtained a distribution of expected values for each voxel, referred to as the reference distribution.

Third, the monitoring PET relative to a specific treatment session scan was simulated and statistically analysed by comparing its voxel values to the aforementioned reference distributions of grey values. The objective was to determine whether either they satisfy the null hypothesis to belong to the simulated voxel distribution or they are attributable to a change in the patient morphology. A confidence level (p-value) was assigned to each voxel. The voxels that were assigned to a p-value lower than the 0.05 threshold identify anatomical regions where a morphological change has occurred.

To quantify to sensitivity of the method, we also simulated a number of intermediate PETs. The latter were derived from synthetic intermediate CT scans, obtained by means of a purposely-built image processing strategy devoted to mimic a gradual change in the patient anatomy.

Fourth, a color-map that highlights the areas in the PET that are reducible to a morphological change was then generated. The map can be superimposed on the CT and can be easily visualized and interpreted by a clinician.

The implementation and testing of the entire approach required a redesign of the simulation framework, originally developed within the INSIDE project. Improvements devoted to a more automatable approach, to allow for the simulation of a large number of PETs on the Pisa INFN data center, have been carried out.

The statistical comparison between the PET of a specific session and the baseline reference PET distribution allowed us to identify and quantify, at the voxel level, the spatial locations where morphological changes occurred. Based on the MC simulations of one patient, it was concluded that visualization and location of morphological changes is feasible, if significant anatomical changes occurred.