Ruijiang Li, Ph.D. - US grants

DESCRIPTION (provided by applicant): Interfraction anatomic changes and intrafraction respiratory motion are the major limiting factors for escalating radiation dose and improving local control in lung cancer radiotherapy. The advent of on-board x-ray imaging device mounted on the medical linear accelerator (LINAC) has provided a tool to obtain valuable anatomic information of the patient in the treatment position. However, due to the slow rotating nature of the on-board imaging system (~1 min per rotation), obtaining volumetric information in real time is extremely challenging. Existing methods have relied on grouping many projections acquired over multiple breathing cycles for several minutes to reconstruct one static anatomy. Further, due to the fact that lung cancer patients tend to breathe irregularly, the reconstructed images are often heavily contaminated by breathing motion artifacts. The goal of this research project is to develop innovative real-time volumetric imaging methods that are able to reconstruct the dynamic patient anatomy in real time (~0.1 s) using a single x-ray projection during dose delivery. This bold goal is made practical by three integral components: effective use of an accurate patient-specific lung motion model, advanced compressed sensing techniques for image reconstruction, and a massively parallel and yet affordable computing platform based on graphics processing units (GPU). During the mentored K99 phase, the candidate will draw on his signal processing and statistical modeling expertise to improve and optimize the patient-specific lung motion model while gaining knowledge in lung patient anatomy and pathology, and to quantitatively evaluate the lung motion model and interpret the clinical significance of the results. During the independent R00 phase, a real-time volumetric imaging method which captures both interfraction anatomical changes and intrafraction breathing motion, will be developed, implemented, and evaluated through systematic phantom and patient studies. Successful completion of this project will overcome a critical barrier to the urgently needed real-time volumetric image guidance in lung cancer radiotherapy and afford a powerful way for us to safely escalate the radiation dose and improve local control of lung cancer. This project fits perfectly with the candidate's long-term career goal of establishing a high-quality independent research program to develop state-of-the-art x-ray imaging techniques, which will provide real-time image guidance for cancer radiotherapy and ultimately improve the therapeutic ratio and enhance the quality of life for cancer patients. Career development and research training will be an integral component during the mentored phase of this project. This training will be further supplemented with formal coursework at Stanford University School of Medicine, as well as participation in research seminars and scientific meetings. The training and research contributions supported by this K99/R00 award will substantially enhance the candidate's career and serve to establish him as a successful independent investigator in the near future. PUBLIC HEALTH RELEVANCE: This project aims to develop real-time volumetric imaging methods that are able to reconstruct the real-time dynamic patient anatomy using a single x-ray projection during dose delivery. Successful completion of the project will overcome a critical barrier to the urgently needed real-time volumetric image guidance in lung cancer radiotherapy. It will provide a critically needed means to treat lung cancer and afford a powerful way for us to safely escalate the radiation dose and improve local control of lung cancer.

? DESCRIPTION (provided by applicant): CT is currently the gold standard in radiation therapy treatment planning. MRI provides a number of advantages over CT, including improved accuracy of target delineation, reduced radiation exposure, and simplified clinical workflow. There are two major technical hurdles that are impeding the clinical adoption of MRI-based radiation treatment planning: (1) geometric distortion, and (2) lack of electron density information. The goal of this project is to develop novel image analysis and computational tools to enable MRI-based radiation treatment planning. We hypothesize that accurate patient geometry and electron density information can be derived from MRI if the appropriate MR image acquisition, reconstruction, and analysis methods are applied. In Aim 1, we will improve the geometric accuracy of MRI by minimizing system-level and patient- specific distortions. To maintain sufficient system-level accuracy, we will perform comprehensive machine- specific calibrations and ongoing quality assurance procedures. To correct patient-induced distortions, we will develop novel computational tools to derive a detailed magnetic field distortion map based on physical principles, which is used to correct susceptibility-induced spatial distortions. In Aim 2, we will develop a unifying Bayesian method for quantitative electron density mapping, by combining the complementary intensity and geometry information. By utilizing multiple patient atlases and panoramic, multi-parametric MRI with differential contrast, we will apply machine learning techniques to encode the information given by intensity and geometry into two conditional probability density functions. These will be combined into one unifying posterior probability density function, which provides the optimal electron density on a continuous scale. In Aim 3, we will clinically evaluate the geometric and dosimetric accuracy of MRI for treatment planning in terms of 3 primary end points: (1) organ contours, (2) patient setup based on reference images, and (3) 3D dose distributions (both photon and proton), using CT as the ground truth. These evaluations will be conducted through patient studies at multiple disease sites, including brain, head and neck, and prostate. Success of the project will afford distortion-free MRI with reliable, quantitative electron density information. This will pave the way for MRI-based radiation treatment planning, leading to an improved accuracy in the overall radiation therapy process. It will streamline the treatment workflow for the MRI-guided radiation delivery systems under active development. With minimal modification, the proposed techniques can be applied to MR-based PET attenuation correction in PET/MR imaging. More broadly, the unifying Bayesian formalism can be used to improve current imaging biomarkers by integrating a wide variety of disparate information including anatomical and functional imaging such as perfusion/diffusion-weighted imaging and MR spectroscopic imaging. It will facilitate the incorporation of multimodality MRI into the entire process of cancer management: diagnosis, staging, radiation treatment planning, and treatment response assessment.

ABSTRACT The goal of this research is to develop and validate prognostic imaging biomarkers for breast cancer. A major challenge in the management of breast cancer is distinguishing patients with indolent disease from those with aggressive lethal disease at diagnosis. Currently, there are no reliable biomarkers to distinguish these groups on an individual level. Consequently, all patients with breast cancer receive adjuvant therapies, but not all benefit equally. This one-size-fits-all approach causes overtreatment, leading to morbidity and mortality. The need for reliable biomarkers is highlighted by the randomized TAILORx trial, which identified a small group of low-risk breast cancer patients who had very low rates of recurrence without chemotherapy, based on the 21-gene Oncotype Dx assay. Unfortunately, a majority (67%) of patients fell in the intermediate-risk range according to the genomic assay, and uncertainty still remains regarding the need for chemotherapy among these patients. Clearly, better biomarkers are needed to improve prognostication and patient stratification in breast cancer. Built on extensive preliminary data, we hypothesize that imaging characteristics reflect underlying tumor pathophysiology, and that image-based phenotyping of both tumor and parenchyma will provide much improved accuracy for recurrence prediction. To test this hypothesis, we propose to: (1) develop and improve methods to explicitly quantify multiregional MRI phenotypes including those of intratumoral subregion and parenchyma, and systematically assess their reproducibility; (2) develop a prognostic imaging signature using a large retrospective cohort of >1000 patients curated by the Stanford Oncoshare Project, and validate it in the prospective multi-center I-SPY 1 cohort; (3) construct a radiogenomic signature to perform additional testing of its prognostic value in 13 public gene expression cohorts of >5000 breast cancer patients. To further improve prognostication, we will build a multifactorial model that integrates imaging with clinical and genomic markers. This research will advance the quantitative imaging field by moving beyond traditional gross-tumor features and incorporating additional parenchymal and intratumoral imaging characteristics. If successful, it will provide much needed, rigorously validated imaging biomarkers for breast cancer, which can be further tested for clinical utility in prospective trials. Ultimately, such biomarkers can be used to stratify patients and guide individualized therapy, by allowing clinicians to avoid overtreatment of indolent disease and intensify treatment in women with aggressive disease.

ABSTRACT Non-small cell lung cancer (NSCLC) is a major disease burden in the United States and worldwide. Most patients are diagnosed at an advanced stage. For unresectable locally advanced NSCLC, the standard of care is definitive concurrent chemoradiotherapy. Unfortunately, the majority of patients will develop local-regional or distant failure with standard treatment. High-dose radiotherapy or consolidation chemotherapy may reduce local or distant recurrence, but are also associated with significant toxicity leading to morbidity and even mortality. Several randomized phase III trials failed to show a survival benefit with intensified treatment given to unselected, locally advanced NSCLC populations, highlighting the limitations of current `one-size-fits-all' treatment. A biomarker-driven approach would allow rational treatment selection based on individualized assessment of risks of local-regional versus distant failure. However, current imaging and genomic markers lack sufficient accuracy in predicting relevant outcomes. The goal of this project is to develop and validate quantitative imaging biomarkers to evaluate early response and integrate with circulating tumor DNA analysis to predict patterns of treatment failure in locally advanced NSCLC. Previously, we developed a novel tumor partitioning method based on FDG-PET and CT images, which revealed spatially distinct tumor subregions with predictive significance in NSCLC. In this project, we will further improve our tumor partitioning method to identify robust subregions, and propose novel image features to characterize intratumoral spatial heterogeneity via spatially explicit analysis. A rigorous qualification procedure will be employed to identify repeatable and reproducible image features for biomarker discovery. We will develop a predictive imaging biomarker by incorporating pre and mid-treatment scans in a retrospective patient cohort, and independently test it in two prospectively collected cohorts including a national randomized phase II trial. Finally, we will combine imaging with circulating tumor DNA analysis in a unifying model to further improve predictive accuracy. We anticipate that the integrated biomarker will allow reliable, early prediction of local-regional vs distant failure, which has important implications for deciding treatment between high-dose RT vs intensive systemic therapy. If successful, the proposed biomarkers will afford a rational approach to individualized therapy and ultimately improve outcomes in locally advanced NSCLC.