NIH grant support
Details about each funded project are given below
The overall goals are (a) understanding relative clinical, biological and immuno-suppressive determinants of response to proton vs. photon therapy; (b) developing personalized response predictive models based on such determinants; and (c) applying individualized (as opposed to population-based) approaches for the selection of the optimum radiation modality for each patient and to enhance the therapeutic potential with IMRT and IMPT dose distributions tailored to an individual patient’s baseline and tumor characteristics. These goals will be achieved in three synergistic projects. Project 1: Understanding Normal Tissue Toxicity to Identify Patients Most Likely to Benefit from Proton vs. Photon Therapy; Project 2: Radiation-Induced Lymphopenia: Understanding, Predictive Modeling and Developing Photon and Proton-Based Mitigation Strategies; and Project 3: Investigating Enhanced Sensitivity of Tumors to Proton Beam Therapy: Mechanisms and Biomarkers. The projects are highly integrated in that treatment modality selection and dose distribution optimization to maximally enhance the therapeutic ratio must consider and balance normal tissue complications (Project 1), radiation-induced immuno-suppression (Project 2) and tumor response based on genotypic factors (Project 3), which cannot be accomplished by any one project alone.
While adaptive therapy has been studied before, none of the previous studies deals with the unique challenges (range uncertainties) and unique capabilities (beamlet optimization and prompt gamma imaging) of proton therapy. This proposal will for the first time address these aspects by developing innovative hardware and software methodologies. We envision treatment planning and delivery to be fully adaptive in terms of intra- fractional changes in patient geometry. This proposal aims at predicting the dose distribution (or a surrogate thereof) in the patient immediately prior to treatment delivery and correct for any discrepancies between the measured and intended dose in less than 2 minutes. This will enable us to deliver (proton) radiation therapy in an adaptive setting and much reduced target volume margins (2mm isotropic plus 2mm range margin in proton therapy in the beam direction) daily while the patient is positioned on the treatment table. We propose to achieve this goal by simultaneously developing fast hardware and software tools that take advantage of in-room prompt gamma and cone-beam CT imaging in combination with fast dose calculation. We will combine this technology with a novel framework on beamlet adaptation. While some of our methods will improve photon therapy as well, we will focus on proton therapy because it offers unique opportunities to dose verification in vivo as well as unique challenges due to range uncertainties. Our methodology will be made available to the entire proton therapy community.
NIBIB R01 EB031102 Constrained Disentanglement (CODE) Network for CT Metal Artifact Reduction in Radiation Therapy
A majority of cancer patients receive radiation therapy as a critical part of their treatment, and many of these patients have metal objects leading to metal-induced CT artifacts. Metal artifacts compromise or preclude radiation therapy in an estimated 15% of all radiation therapy patients. Our overall goal is to eliminate CT metal artifacts in general and improve RT in particular using an integrated deep learning solution.
NCI R21 CA252562 Understanding the importance of dose-rate variations for patient treatments in FLASH and conventional radiation therapy
Radiation therapy has been an important component in the treatment of cancer for many decades. While the parameters of tumor control have been relatively well understood, normal tissue complication probabilities (NTCPs) are often only thought of as constraining factors in the design of a treatment regimen. However, with continuously improving treatments of primary lesions, NTCP and their impact on the quality of life have moved into the focus of research. One of the most prominent methods reported to spare healthy tissue is FLASH radiation therapy, i.e. ultrafast (>40 Gy/s) irradiations, which reportedly results in incredible tissue sparing effects without compromising efficacy for tumor cure. Conversely, very low dose-rate irradiations have also been reported to offer protective features. Together, these experiments suggest that the current clinical routine, treating patients with a medium dose-rate of ~2 Gy/min, could in fact be the least favorable when considering healthy tissue side effects. In clinical practice, patients are treated with a variety of radiation treatment modalities. While the resulting target doses are similar across modalities, the dose delivered to the normal tissue and the time structure of the delivery can vary significantly. Despite the range of dose-rates across treatment plans, the biological effect is estimated only based on the total dose received. We hypothesize that dose-rate plays an important role in the outcome of radiation therapy that can be exploited for specific treatment scenarios. For extreme cases such as FLASH therapy, averaged dose-rates do not capture the relevant time structures adequately. We will apply Monte Carlo simulations to determine time structures from the clinical scale of minutes (spot scanning / gantry rotation time) to nanoseconds (intra-spot delivery time) and assess potential effects on healthy tissue sparing that would improve the quality of life for radiation therapy patients.
NCI R21 CA279068 GPU-based SPECT Reconstruction Using Reverse Monte Carlo Simulations
Interest in applications of radiopharmaceutical conjugates has notably increased in the last few years for the treatment of a variety of cancers. These conjugates are composed of chelators to target cancer cells and radionuclides to employ the cytotoxicity of ionizing radiation. Radiation dosimetry is required to determine the dosages, efficacy, and safety of these treatments, and 3D quantitative imaging of the biodistribution of activity represents the best tool to perform dosimetry. For most radionuclides employed (non-positron-emitters), SPECT imaging is needed for patient-specific dosimetry. However, multiple physical factors affect SPECT image quality, such as attenuation, scattering, or the response collimator-detector system in SPECT scans. To account for them, Monte Carlo techniques can be used due to their remarkable accuracy in representing physical processes relevant to the transport of ionizing radiation. In particular, 3D SPECT reconstruction from the acquired bidimensional projections may be obtained by transporting backward the photons detected in the gamma camera projections, although many photons to be reversely transported require specially optimized architecture and simulations. This project will develop a new reverse Monte Carlo software for SPECT reconstruction, built from scratch in the GPU to speed up simulations. First, only the relevant reverse physical processes will be selected and modeled using inverse processes of the well-characterized TOPAS Monte Carlo code for radiation transport. Then, specific properties of collimator-detector systems will be integrated into our code to determine the angular distributions for the photons detected. Finally, these developments will be integrated into a GPU-based platform and shared with the Informatics Technology for Cancer Research program at NCI for further results of specific commercial SPECT scans from the research community.
NCI contract 2020A003480 Monte Carlo for Pediatric Proton Epidemiology