Announcements IV: January 23, 2008

1) Wednesday, January 23rd, at noon; Alway M104
Julia Patriarche, PhD
Mayo Clinic

Title:
"Detection of Change in Serial Magnetic Resonance Studies of Brain Tumor Patients"

Abstract:
The comparison of serial magnetic resonance imaging studies is a common task in clinical radiology. It is, however, widely considered not to be very reproducible. There are a variety of reasons for this, including the confounding of disease-related changes with acquisition-related changes and issues related to information presentation. We have constructed a computational system that performs the comparison of serial magnetic resonance imaging studies and presents changes in the form of a color-coded change map, superimposed on the anatomical images. The system additionally formats the output as a quantitative summary. We used this quantitative summary to conduct a study with 88 brain tumor serial comparisons. Our results were suggestive that it may be possible to use the change detector to identify cancer changes months earlier than is possible using manual inspection, alone.

We have recently implemented an integrated system for the change detector, which includes a graphical user interface (GUI). The GUI not only displays the color-coded change map, but also allows the user to turn it on and off. The GUI provides linked cursors, and it additionally provides "flicker" functionality to allow the user to rapidly alternate between the serial acquisitions. We are preparing to deploy the GUI change detector clinically, which will greatly increase the size and variety of possible future research studies and which will allow the direct clinical application of this technology.

The change detector is an example of a layered artificial intelligence (AI) architecture in which each layer builds upon the layer below, with each layer accomplishing progressively more sophisticated analyses. Specifically, the change detector is built on a lesion-finder application. The lesion finder is built on an automated sample point's algorithm. The automated sample point's algorithm is built on a significant region detection algorithm. Each of these algorithms has merit in its own right, and each can be used in a modular fashion in a variety of contexts. As a unified application, they together automatically address a complex clinical task. Early detection of changes may facilitate improved care through more rapid intervention following recurrence. It may also facilitate screening and personalized therapy. We additionally see the change detector as providing a solution to the problem of novel therapy comparison, by providing fully automatic, reproducible, and quantitative measures of change. We envision the change detector as a model of layered artificial intelligence, not only freeing the radiologist from the drudgery of information overload, but providing a model whereby greater information will enable many sophisticated automatic analyses by the computer, with the computer bringing to the attention of the clinician only what is relevant.

Biography:
Julia Patriarche is an informatics fellow in the Radiology Informatics Lab at the Mayo Clinic College of Medicine. She has completed an undergraduate degree in electrical engineering/computer engineering option at Queen's University in Kingston, Canada; a PhD in medical science/medical imaging; and a neurology fellowship at the Mayo Clinic College of Medicine.

2) Thursday, January 24th, at noon; Alway M112
Ross Mitchell, PhD
University of Calgary

Title:
"Virtual Biopsies: Non-Invasive Molecular Diagnosis"

Abstract:
Our expanding knowledge of the genetic basis and molecular mechanisms of cancer is beginning to revolutionize the practice of clinical oncology. Increasingly, molecular biomarkers of prognosis and treatment response are being used to classify tumors and direct treatment decisions. Advanced medical imaging platforms such as MRI, PET, and CT provide incredibly detailed images of tumors that reflect their structure, biochemistry, physiology, and perhaps genetics.

Studies by the Imaging Informatics Lab at the University of Calgary, and others, show that information about a tumor's molecular phenotype can be obtained by using novel algorithms and computational tools to more fully analyze tumor images. Such "virtual biopsies," performed by applying these image-processing techniques to routine diagnostic images (e.g. MRI, PET, or CT), could be a rapid and powerful means of assaying important cancer biomarkers. If successfully validated, and proven to have suitable sensitivity and specificity, the use of non-invasive, imaging-based molecular diagnostic tests would offer significant advantages over conventional surgical biopsies. For example, this could be important in the context of large heterogeneous tumors, multiple metastases, surgically inaccessible tumors, and settings where disease progression needs to be monitored frequently over time. Virtual biopsy research lies at the intersection of molecular imaging, medical imaging physics, and biocomputation, and is highly complementary to these areas. This presentation will cover key enabling technologies behind virtual biopsies and discuss some recent successes in this research.

Biography:
Dr. Ross Mitchell is an associate professor of the Departments of Radiology and Clinical Neurosciences and an adjunct professor of the Department of Computer Science at the University of Calgary. He is also the founding and chief scientist of Calgary Scientific Incorporated, a Multiple Sclerosis Society of Canada; a Donald Paty Scholar; and an Alberta Heritage Foundation for Medical Research Senior Scholar. Dr. Mitchell has received numerous awards for his research including the Berlex Canada MS Research Award; several Dean's Awards of Excellence from the University of Western Ontario; Best Paper Awards from the Canadian Association of Radiologists and the International Organization for Medical Physics; and two Awards of Merit from the Radiological Society of North America. Dr. Mitchell has a proven research track-record comprising 11 patents, 73 invited presentations, 63 peer-reviewed articles, and 150 published abstracts.

Dr. Mitchell supervises a research team investigating space/frequency analysis, medical image processing, as well as segmentation and visualization technologies. For more information, please see, http://www.ImagingInformatics.ca.

3) Monday, January 28th, at noon; Alway M104
Jianming Liang, PhD
Siemens Medical Solutions USA Inc., Malvern, PA

Title:
"Dynamic Chest Image Analysis, United Snakes, and
Computer-Aided Detection"

Abstract:
Modern medical imaging systems generate enormous datasets with ever higher coverage and resolution, but it is the clinically relevant information in these images that is paramount. I shall present several novel computational approaches for gleaning such information from chest X-ray images to reveal pulmonary functional abnormalities, for segmenting and characterizing organ motions, and for detecting the most lethal diseases from CT images, including pulmonary embolism and colonic polyps. The former approach has yielded model-based analysis and visualization methods for revealing focal and general abnormalities of lung ventilation and perfusion based on a sequence of digital chest fluoroscopy frames collected with the dynamic pulmonary imaging (DPI) technique.

In particular, I shall present a novel multiresolutional method with an explicit ventilation/perfusion analysis model, as well as "United Snakes," an interactive deformable model framework for lung registration and motion analysis, cardiac shape and motion analysis, and other applications. Finally, I will introduce a fast yet effective concentration-oriented tobogganing technique for efficient local artery/vein separation and multiple instance classification for the automated detection of pulmonary embolism from CT pulmonary angiography (CTPA), and a virtual colonoscopy technique that simplifies the complex 3D-polyp detection problem into a 2D-disk identification problem, significantly improving sensitivity while reducing computation time.

Biography:
Dr. Jianming Liang is a staff scientist at Siemens Medical Solutions USA, Inc., where he has been engaged in research and development activities in the domain of computer-aided diagnosis in medical imaging since December 2002. He holds a PhD degree (2001) in computer science and carried out his thesis work at the Turku Centre for Computer Science in Finland and in the Visual Modeling Group at the University of Toronto in Canada. From 2001-02, he was a Natural Sciences and Engineering Research Council (NSERC) of Canada Industrial Research Fellow. His research on dynamic chest image analysis received a University Faculty Research Award from the University of Turku. His other prizes include a Siemens Recognition Award and a Best Paper Award at the 2007 International Congress of Computer Assisted Radiology and Surgery in Berlin, Germany.

4) Thursday, January 31st, at noon; Clark Center Auditorium
Daniel Rubin, MS, MD
Stanford University

Title:
"Imaging Informatics: From Bench to Bedside and Beyond"

Abstract:
Vast amounts of knowledge lie within the grasp of radiology researchers and practitioners to help them to understand disease and to practice effectively, but much current biomedical knowledge is not being accessed and utilized. The explosion in images and non-imaging data is challenging the ability of radiology researchers to identify and to pursue promising new investigational directions. The latest results that researchers produce are not always informing radiologists in their day-to-day work, as there are few tools to help them to identify, retrieve, and use pertinent clinical and research knowledge at the point of care. Consequently, there is variability among radiologists in their clinical effectiveness, and opportunities for translating new discoveries into practice are being lost. The methods and tools of biomedical informatics are enabling biologists to cope with similar problems arising from the information explosion in biology, and they are adopting informatics techniques to function effectively in the e-Science era.

In this presentation, I will discuss ongoing work to develop and apply biomedical informatics techniques to meet the information challenges in radiology. Specifically, knowledge representation, semantic annotation, statistical natural language processing, data integration/warehousing, computer reasoning, and decision support are key directions in informatics needed to create intelligent applications for radiology. Future advances in radiology will lie at the intersection of imaging science and biomedical informatics. The new computer applications that emerge will change clinical imaging workstations into knowledge portals and enable radiologists to keep pace with new discoveries, to exploit new radiology knowledge, and to practice more consistently and effectively.

Biography:
Daniel Rubin is a research scientist in the Center for Biomedical Informatics Research and clinical assistant professor of radiology at Stanford University. He is director of scientific development for the National Center for Biomedical Ontology, a National Center for Biomedical Computing of the NIH Roadmap. He is chair of the RadLex Steering Committee of RSNA, chair of the Informatics Committee of the American College of Radiology Imaging Network (ACRIN), and co-chair of the Medical Imaging Systems Working Group of the American Medical Informatics Association. In addition to informatics, his background includes clinical and investigational radiology, as a radiologist and researcher. His academic focus is the intersection of biomedical informatics and imaging science, developing computational methods and applications to access and integrate diverse clinical and imaging data, to extract information and meaning from images, to enable data mining and discovery of image biomarkers, and to translate these methods into practice by creating computer applications that will improve diagnostic accuracy and clinical effectiveness.