The Academy's 2019 CECI2 Class

My research program has developed imaging measures that characterize the brain tumor environment. Brain tumor cells influence other non-tumor cells around them to help them thrive. My research has furthered the understanding of these interactions and the effects of therapy upon the brain tumor. As a result, we can now better determine who will survive longer. This is particularly important for new therapies that act to unleash the patient’s immune system to fight the brain tumor. Using standard of care imaging methods, we cannot determine who will respond to immunotherapy. My lab has recently developed a new approach of imaging these immune cells in a way that is easily done in patients and captures the effectiveness of therapy. I hope to discuss the importance of these new imaging measures and the necessity of their use in clinical trials to specifically capture if the therapy is working.



I’m also interested in developing methods to measure inflammation in skeletal muscle via 19F and 1H MRI, which will allow for early detection of both pathologic changes in many disease states and could help to identify early responses to therapy. Finally, I’m interested in studying blood flow and metabolism in muscle to better understand the ramifications of microvascular dysfunction in diabetes and cardiovascular disease.

I am interested primarily in the diagnosis and treatment of cancers using radioactivity in the field of Theranostics. For example, we currently use a molecule which binds to a cell receptor which is overexpressed in a specific type of tumor known as a neuroendocrine tumor. We are able to label this molecule with a radioactive isotope which can be injected in the patient, travels to the specific cell type, and used to image the patient to see if the receptor for the molecule is being overexpressed in the tumor for which we are looking.
If we see the tumor on the images, we can remove the imaging radioisotope from the molecule, and replace it with a high energy isotope, which is not used for imaging, but is used for treatment.
This high energy isotope can only travel a few millimeters in tissue, so what happens is the molecule with the high energy is injected into the patient, the molecule travels through the body until it finds the tumor and binds to the overexpressed receptor. After it binds the receptors on the tumor, it releases the high energy radioactivity which can only travel a few millimeters, thus destroying the tumor without damaging the surrounding healthy tissues.
This technology is currently only FDA approved for use in rare neuroendocrine tumors and for treating thyroid cancer, and, as one could imagine, there are many other tumors we could use this technology to target.

The goal of my work is to include the effects of these real-world problems into our understanding of MRI. This allows us to produce higher quality diagnostic images and makes MRI accessible to a wider range of patients (such as children and the elderly, who are likelier to move during MRI exams). However, adding in real-world effects presents large computational problems, taking days to run even on state-of-the-art computers. I use advanced mathematical techniques and artificial intelligence to break down these large problems, so they can instead take minutes or seconds to solve, allowing their use in a wide variety of clinical settings.



In another research direction, we are developing computational methods to obtain biomarkers from medical images with the goal of personalizing medicine in patients with cancer. In one project, we observed in a prospective clinical trial that a set of machine-learning-based methods we had developed to quantify metrics from diffusion MR images were important for predicting therapy response at an early stage in patients with liver metastasis. This is highly significant since early prediction can help adapt the therapy regimens for the patients, thus helping to reduce side-effects and therapy costs, and potentially yield improved clinical outcomes. Similarly, we recently developed artificial-intelligence (AI)-based methods to quantify metrics from positron emission tomography (PET) images of patients with lung cancer, again with the goal of predicting therapy response at an early stage and thus adapting therapy. Research along this direction will be the focus of my presentation at the Medical Imaging Technology Showcase.


One potential topics of interest to discuss at congressional meetings is my ongoing research on the use of patient navigation programs to reduce barriers and disparities in access to breast imaging. Another is my work on optimizing imaging evaluation of women with signs and symptoms of breast cancer, both in the United States and internationally through global cancer collaborations. I have also led multi-institutional collaborative studies on the appropriateness of breast MRI reporting practices and diagnostic mammography reporting practices.
Prior to my clinical research in breast cancer imaging, much of my earlier research experiences were in the laboratory sciences in the field of molecular biology. This basic science research background helped foster my skills in scientific methodology and developed a strong foundation for my current clinical-translational research.
The basic design of the conventional x-ray tube has not significantly changed since its invention by William Coolidge in 1913. Almost all current diagnostic x-ray tubes utilize a heated tungsten filament to generate the electron stream necessary for x-ray production, which limits the ability to put multiple x-ray sources in close proximity and reduces the precise control of the source. A new x-ray tube based on carbon nanotubes was invented at the University of North Carolina at Chapel Hill in 2003. My research is focused on bringing x-ray sources based on this tecnology to a variety of clinical applications, including 3-D breast imaging, cardiac, chest and orthopedic imaging. The new x-ray source has the potential to reduce the radiation dose to patietns, provide higher quality images, and be used to develop portable imaging devices for the military and emergency medical services. The CNT sources serve as an example of the complexities in translating and adapting a novel imaging technology to the clinic.
My area of interest is understanding of the effects of lung diseases on the heart and the vessels in the body. I am interested in exploring the role of diagnostic tests (CT and MRI) to further our knowledge and eventually use that understanding to decrease the burden of cardiovascular death. Cardiovascular disease has been the most common cause of death for > 50 years. While smoking is declining with an aggressive campaign regarding its harmful effects, the use of e-cigarettes is on the rise, especially among young adults. The focus of my current research is to study the effects of e-cigarettes use on the heart and compare it with traditional smoking. To further our understanding of the disease, I am involved in the development of MRI sequences that enables its use without the need for breath-holding. Because of a need for breath-holding, many patients that need cardiac MRI cannot get the test. With such development, patients that have advanced lung disease can also be studied to understand the effects on the heart using imaging.


In these research projects, I have applied advance imaging techniques and postprocessing analysis to examine the details of brain connectivity in children with neurodevelopmental disorders, including autism. We have used a special type of MRI (called DTI) to assess nerve fibers as conduits of connectivity in the brain at a microstructural level. We have also used “functional” MRI studies to look at synchronicity of fluctuations in the blood oxygen levels at different areas of the brain as a measure of activity pairing – or “functional connectivity”. Together these data provide complementary, but independent measures of brain organization and connectivity – i.e. Connectome. In the next step of our research, we will apply machine learning models to incorporate information from both these MRI techniques into one framework to devise novel quantitative biomarkers, which could guide risk stratification, prognostication, and treatment planning in children with autism.
• Automated Assessment of Stroke on CT Scan using Artificial Intelligence
Availability of mega-data and machine-learning algorithms have paved the road for development of novel diagnostic and prognostic solutions in medical imaging. The Yale New Haven Health system is the largest health care delivery system in Connecticut, including four hospitals (Yale New Haven, Bridgeport, Greenwich, Lawrence &Memorial) with 2409 beds and over two million inpatient admissions per year. The Yale Stroke Center is the lead hub of the Southern New England Partnership in Stroke Research, Innovation and Treatment (SPIRIT) of the NIH StrokeNet with approximately 1,800-2,000 code stroke per year. The centralized Image Archiving System and Electronic Medical Record system have provided a unique opportunity for collecting the mega-data necessary for machine learning models.
Our research group is applying artificial intelligence algorithms for automated detection of stroke on head CT scan and measuring the amount of tissue damage surrounding a brain bleed. Such automated tools can help with triage and diagnosis of critical findings such as acute stroke, which may be particularly important in resource-challenged parts of the country such as smaller, rural communities without access to subspecialty trained radiologists.
• Risk Stratification and Treatment Planning for Patients with Brain and Neck Tumor using Machine-Learning Quantitative Assessment of Medical Images
In my research projects, I have used “radiomics” analysis to extract and apply textural features of brain and neck tumors for diagnosis and prognosis prediction. “Radiomics” refers to computerized assessment of medical images for extraction of features that are often hidden from human eyes. With this technique, we can extract thousands of features and data points from CT, MRI, or PET studies and apply these information to devise quantitative and objective prognostic biomarkers, which can improve outcome prediction, and treatment planning in cancer patients. Machine-learning algorithms can combine such radiomics features with anatomical findings and clinical data to develop survival models for prognostic classification of patients with brain or neck cancer. Such accurate risk stratification and prognostic staging are the first steps toward achieving “personalized” treatment decisions and “precision medicine” practice in this field.


In another research direction, we are developing computational methods to obtain biomarkers from medical images with the goal of personalizing medicine in patients with cancer. In one project, we observed in a prospective clinical trial that a set of machine-learning-based methods we had developed to quantify metrics from diffusion MR images were important for predicting therapy response at an early stage in patients with liver metastasis. This is highly significant since early prediction can help adapt the therapy regimens for the patients, thus helping to reduce side-effects and therapy costs, and potentially yield improved clinical outcomes. Similarly, we recently developed artificial-intelligence (AI)-based methods to quantify metrics from positron emission tomography (PET) images of patients with lung cancer, again with the goal of predicting therapy response at an early stage and thus adapting therapy. Research along this direction will be the focus of my presentation at the Medical Imaging Technology Showcase.


Potential topics include:
– What the current standard of care for female health evaluation is,
– Limitations of the current standard of care,
– Physiological changes of the female body that prevent access to the standard of care at different stages in life,
– Information lacking in the literature needed for the improvement of women and fetal care,
– The role of quantitative MRI in improving women’s healthcare.
– MRI screening and staging of female pelvic cancers and breast cancer

In the past, as part of a research team I investigated the use of amino acid analog positron emission tomography (PET) radiotracer fluciclovine (FACBC, Axumin) in different cancers, such as prostate, breast, and lung cancers. These investigations led to the Food and Drug Administration (FDA) approval of fluciclovine for prostate cancer imaging.
Currently, as a PI, I received a small grant to investigate the feasibility of FACBC PET radiotracer in the use of gynecological cancers, such as ovarian, endometrial, and cervical cancers. Preliminary data are promising. My goal is to develop prospective protocols for each of the above-mentioned cancers to investigate the ability of fluciclovine PET tracer to localize areas of malignancy. The rationale is that up until recently F18-Fluorodeoxyglucose (FDG) tracer was the only FDA approved tracer for PET imaging. FDG enters the cell via glucose transporters (GLUT). GLUT are highly expressed in cancer and inflammatory cells which may decrease its ability to localize deferent cancers. FDG normally washes out from the body by the kidneys which results in an intense urine accumulation. This may obscure areas of cancer involvement in the pelvis. Therefore, there is a need for a better biomarker for accurate imaging. Fluciclovine is a synthetic amino acid PET tracer, proven to successfully localize small prostate cancer lesions. As well, fluciclovine uptake is not very high in inflammatory cells and has a very slow renal washout, resulting in a clear view of the pelvis. Fluciclovine is not a prostate cancer cell-specific tracer and will enter the cell by the amino acid transporters. Amino acid transporters are overexpressed in ovarian, endometrial and cervical cancer cells. Thus, we believe that fluciclovine will enter these cells in higher concentration compared with benign cells.



STRATEGY & GOAL: The goal of our research is to validate novel imaging techniques for detection of plaques that are risk of complication through detection of underlying molecular and cellular events leading to plaque rupture. We are specifically focused on detection of inflammation within the vessel wall, as a key factor in pathogenesis of atherosclerosis and its complications.

The topic of my presentation at the Medical Imaging Technology Showcase will be functional brain connectivity alterations in patients experiencing persistent vestibular symptoms (dizziness and vertigo) after concussion using task-based and resting state functional MRI.
