Machine-learning models and shielding materials for radiation protection of patients and personnel in ion therapy

Compared to traditional external X-ray therapy, ion therapy uses heavy ions to improve dose conformality, increase biological effectiveness, and spare the healthy tissues surrounding the tumor volume. With these unique physical characteristics, ion therapy can maintain or even escalate the dose to the tumor to achieve greater cancer control while reducing severe toxicity risks. Despite these potential advantages, the widespread adoption of ion therapy still faces some technological challenges pertaining to the radiation protection of patients and personnel, including 1) the effective implementation of adaptive treatment workflow to account for anatomy changes, 2) optimal tumor coverage with the therapeutic beam to minimize the stray radiation delivered to healthy tissues, and 3) shielding of highly-penetrating secondary radiation. The work described in this dissertation focuses on proton therapy and aims to improve these three aspects of ion therapy and enable a safer and more effective treatment workflow. Radiation treatment is typically split into tens of fractions, delivered at time intervals of several days. Adaptive radiation treatment (ART) repeats the imaging and organ contouring for each treatment session and uses the newest anatomy information to update the treatment plan. However, this process is often impractical due to the long manual contouring time. This dissertation investigates the feasibility of using machine-learning models to perform fast image contouring and aid the adaptive treatment workflow of proton therapy. We developed a three-dimensional (3D) volumetric segmentation model based on V-Net to automatically contour the organs. We significantly improved its performance in terms of computation time by producing sparse yet information-rich point-cloud data from voxel-based data and implementing a point-cloud-based segmentation model for prostate cancer treatment. Our model achieves an 89% average segmentation accuracy of the prostate and a rapid 1.5-second segmentation speed, outperforming the state-of-the-art voxel-based segmentation models. Proton beams exhibit the highest energy deposition, commonly referred to as the Bragg peak, at the end of their path in tissue. The distance that a particle travels through matter is referred to as its range. Accurate verification is needed to confirm the location of the beam range and ensure that the highest ionization density spares healthy tissues surrounding the target. This dissertation investigates the feasibility of using a novel organic scintillator, deuterated stilbene (stilbene-d12), capable of detecting and discriminating both gamma rays and neutrons to perform real-time range verification. After thorough experimental characterization of the stilbene-d12 light-output response, we used Monte Carlo simulation to demonstrate that the fall-off of the neutron flux remotely measured by the stilbene-d12 can be used to locate the beam range reliably. Using the 90% fall-off of the measured neutron flux, the Bragg peak is measured with an uncertainty of ±1 mm (1 standard deviation). Proton beams can produce highly penetrating secondary neutrons and gamma rays during the treatment by interacting with the patient and surrounding structures. Therefore, adequate and specific radiation shielding is needed to protect personnel at proton therapy facilities. This thesis reports the development and testing of novel geopolymer (GP)-based materials for improved radiation shielding. Our results show that tungsten and polyethylene-loaded GP composites achieve attenuation coefficients that are 30% and 23% higher than high-density concrete for gamma rays and fast neutrons, respectively. Moreover, GP composites with boron-based dispersants show 49% improvement in thermal-neutron absorption compared to commercial thermal-neutron shielding materials. Therefore, using the developed GP composites as shielding materials can significantly reduce the construction footprint of ion therapy units.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Nuclear, Plasma, Radiolgc Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Zhou, Jianxin

Contributors dc:contributor

• Di Fulvio, Angela
• Rizwan, Uddin
• Meng, Ling Jian
• Kriven, Waltraud M
• Wang, Yuxiong
• Hermann, Gregory M

• Machine Learning
• Image Segmentation
• Geopolymer
• Neutron Detection
• Ion Therapy

Statement dc:rights

• Copyright 2024 Jianxin Zhou

Language dc:language

eng, en

Handle dc:identifier

https://hdl.handle.net/2142/127335

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/127335