Development and Application of Image Analysis Methods

One of the greatest strengths of the University’s medical imaging program is the degree of active cross-departmental, cross-college, inter-university and international collaboration activities. University researchers from medicine, image processing, pattern recognition, artificial intelligence, biostatistics, mathematics, etc. routinely collaborate on a number of federally and industry-funded translational research projects that rely upon imaging. These collaborations result in new methods for animal and/or human imaging, quantitative analysis and its use in epidemiologic and clinical trials, and effective translation of research results obtained under controlled conditions to clinical medicine.

In each project described below, multi-disciplinary teams of University researchers have merged medical imaging and image analysis with translational outcomes, frequently defining new interdisciplinary fields. These imaging projects demonstrate the wide breadth of cutting-edge interdisciplinary translational research projects currently performed at the University.

3-D and 4-D Coronary Hemodynamics and Local Atherosclerosis (Sonka, PI, 1999-2009, R01 HL 63373). The main goal of this project is to develop a quantitative approach to a comprehensive description of coronary atherosclerosis morphology and associated coronary function in vivo. Image data acquired during routine coronary intervention from biplane angiography and intravascular ultrasound (IVUS) are used to form a computational 4D model of the diseased coronary artery. Identified plaques are classified by four plaque types. The geometrically correct model is subjected to computational fluid modeling to identify hemodynamic shear stresses in the artery. The obtained quantitative information about the coronary artery is used to predict the atherosclerotic status 1 year after the initial intervention (See Figure 2.10-2).

Arterial Endothelial Function, Genetic Causes of Cardiovascular Disease—An Epidemiologic Study. (Burns PI, 1966-2010, R37 HL61857) Vascular Tools software for quantitative analysis of brachial flow-mediated dilatation (FMD) was developed for use in identifying early indicators of the atherosclerotic process (image analysis part of the project funded by NIH, 1999-2010). The developed quantitative analysis methods are used locally to support the “Muscatine Study” – a 35-year-old ongoing study of cardiovascular disease risk factors starting in youth. Over 800 subjects have been enrolled in this study, which started in 1970. The Vascular Tools Brachial Analyzer software has become a de-facto standard for brachial FMD image analysis worldwide and is used in more than 90 research institutions. Genetic data about cohort participants and various phenotyping data going back to 1966, together with a variety of image-based information (coronary calcium assessed by CT, brachial FMD, carotid IMT, aortic IMT), are being studied to identify the ability of genetic + phenotypic information for prediction of cardiovascular disease development. Current results demonstrate that development of cardiovascular disease can be predicted at birth with about 69% accuracy from genetic information only. Adding phenotypic information in adulthood increases the match between cardiovascular disease and quantitative genetic and phenotypic indices to over 84% in a cohort of more than 600 subjects.

Early Predictors of Stroke—Assessment of Carotid and Aortic IMT (Davis, PI, 2004-2008, R01HL054730). This project is closely related with the above described Muscatine Study. Using the same cohort, early predictors of stroke and heart attack are studied by quantitative analysis of intima-media thickness (IMT), arterial compliance, and plaque characterization in carotid arteries and the aorta.

Computer-Aided Diagnosis of Congenital Heart Disease from Cardiovascular MR Data (Sonka, PI, 2003-2007, R01 HL071809). We are developing novel image-based techniques allowing early detection of aortic aneurysm and rupture development in patients suffering from congenital connective tissue disorder. Using annual MR images of a normal and diseased group of subjects, automated segmentation of the aorta in 4-D, yields a number of morphologic and functional indexes describing aortic shape and motion. Based on these indexes, we are attempting to predict near-future changes of the aortic shape and function so that preemptive surgeries can be timed appropriately. Computer-aided diagnostic methods and training from a 4-year temporal sequence of normal and diseased MR data are used to solve the problem. At this point, our results demonstrate the ability of index-based analysis to distinguish between normal and connective-tissue-disorder subjects – a result that cannot be obtained by expert visual analysis of the MR data.

Image- and Model-Based Analysis of Lung Diseases (Hoffman, McLennan, Lin, Thorne, Wenzel, 1999-2011, R01-HL64368, R01 HL64368, U01CA091085, R01 EB005823, R21/R33 EB001689, P30-ES-005-605, R21CA094310, R01HL069174, R01 HL073598-01, PAS-99-010, R01 HL64368). University research in pulmonary imaging and pulmonary image analysis spans several departments and includes many investigators working on several large projects. The group’s major areas of focus include novel image acquisition methods for pulmonary imaging, image segmentation and registration for processing pulmonary images, and the development of computer-aided diagnosis systems for pulmonary CT images.

Figure 2.10-2. Lumen/plaque and media/adventitia border segmentation in an IVUS image sequence: (a) original cross-sectional and long-axis images; (b) segmentation result, showing two borders; (c) 3-D reconstruction of a vessel showing the lumen/plaque surface in red and media/adventitial surface in green (d-h) Morphology/geometry analysis: (d) part of the original angiogram; (e) 3-D model in the same view (note that the left circumflex branch has not been modeled); (f) plaque distribution; (g) local curvature; (h) classification results of the left anterior descending branch distal of the bifurcation.

• Novel Image Acquisition.

  Research in novel image acquisition for pulmonary imaging includes respiratory-gated imaging for studying the breathing lung, perfusion imaging, and ventilation imaging with CT using injected or inhaled contrast agents. Low-dose imaging is another active area of pulmonary imaging research; it will directly translate to clinical care. The newest area of interest is the development of MR approaches based on hyperpolarized-gas MR imaging of the lung.

• Software Development.

  The University group has developed a comprehensive set of software tools for pulmonary CT image processing and analysis. The set comprises software modules for lung, lung lobe, airway, and vessel segmentation, airway tree structural analysis and geometry measurement, parenchymal tissue characterization, pulmonary blood flow measurement, and pulmonary ventilation analysis. Newly developed capacities include computer-aided diagnosis systems for pulmonary nodules, and lung-image registration methods for intra-subject and inter-subject matching. Intra-subject registration has been used to infer pulmonary mechanics (eg, strain) from multiple respiratory-gated images; the inter-subject registrations form the basis of an effort to construct a comprehensive image-based atlas of the normal human lung. Figures 2.10-3 through 2.10-8 are examples of some of the quantitative capabilities developed at the University.

• Human Lung Atlas.

  The group is creating a CT-based atlas of the normal human lung with associated parameters representing anatomic and physiologic ranges of variability for normal human beings. For this project, image registration is being used to combine image data from different individuals and of different modalities into a single coordinate system. The goal is to produce an atlas that is a true representative of the normal population, such that suspected pathologies or abnormalities can be tested by comparing a candidate lung scan with the atlas. This project includes lung images of humans, sheep, and dogs. The studies related to animals are aimed at understanding the lung's regional mechanics and include images at different air pressure levels, pre/post injury scans, etc.

  

  Figure 2.10-3. Lung, lobe and airway segmentation. Right and left lungs are found along with major fissures.

  

  Figure 2.10-4. A. Adaptive region growth-based airway segmentation; B. Extraction of topological centerlines-and geometrically correct thinning; C. Partitioning of segmented tree via isotropic label propagation; D. Mathematical graph representation of the individual tree. The tree shown in D is stored in an XML file providing associated measures for each part of the tree. These measures include segment length, branch angles, regional luminal area, wall thickness, segment luminal volume, segment surface area, regional minimum diameter and the diameter of the orthogonal as well as regional maximum diameter.

  

  Figure 2.10-5. Once the airway is segmented and the centerline is found, we are able to automatically label the segments. This provides both a roadmap to the lung periphery as well as a way to match one subject across time and to match multiple subjects to our atlas to compare the individual to the range of normal. The above image allows interaction by clicking on a segment to get luminal and wall dimensions.

  

  Figure 2.10-6. Assessment of the lung parenchyma based upon the lung density (attenuation) histogram. Upper left panel: Density histogram; Upper middle panel: Accumulation histogram; Upper right panel: Quantitative report based upon the lung density evaluation: Middle row: Coronal, Transverse, Sagittal sections of volumetric CT image of the chest with a color overlay showing the location of the voxels falling below the selected threshold (yellow vertical line in the upper row graphs). Also shown are the regions automatically selected for the core and rind measurements. In these images the lungs are also divided into thirds based upon the apical-to-basal distance measure. With thinner section scanning, the lung can be divided into actual lobes. Lower row: three views of a volume rendering of the lung with the distribution of voxels falling below the selected threshold depicted in green.

  

  Figure 2.10-7. Characterization of regional lung parenchyma based upon the Adaptive Multiple Feature Method (AMFM) with an extension of the features to 3D. EC = Emphysema in subjects with GOLD 2 and 3 COPD; MC = Emphysema in subjects with GOLD1 COPD; NC = normal-appearing lung image obtained from GOLD 1 COPD; NS = Normal-appearing lung image obtained from GOLD 0 smokers.

  

  Figure 2.10-8. Blood flow (Middle Column) and Mean Transit Time (Right Column) maps in a prone sheep followed through the time course before (Upper Row) and following administration of papain to the right lung (Middle Row). Note that on day 16 while the visible evidence in the gray-scale images (Left Column) has largely cleared, the blood flow maps remain as clear evidence of the presence of interstitial pathology. Also note the unexpected finding of the involvement of the dorsal region of the left (untreated) lung. Color scales for the images (blue to red) are 4-12 ml/sec/g tissue and 3-7 sec for PBF and Mean Transit Time, respectively.

Industry Affiliations. The University’s lung imaging group has strong links to industry, including several scanner manufacturers (Siemens, GE, Philips, Toshiba), respiratory product manufactures (Ferraris, Senormedics), drug companies (Novartis, Eli Lilly), and image analysis software vendors (Vital Images, VIDA Diagnostics).

Coronary Angiogenesis from Micro-CT (Tomanek, PI, 2004-2008, R01-HL075446). The pulmonary image analysis research described above was utilized in a coronary angiogenesis project by adapting the intrathoracic tree segmentation and analysis methodology to the analysis of micro-CT images of coronary trees in neonatal normal and genetically engineered mouse hearts (2004-2009, NIH). While image analysis is only a small part of this project, the quantitative indices that could be obtained from 5 to 10 μm, 3-D micro-CT images of radiopaque plastic-perfused coronary trees offered a completely new approach to studying angiogenesis in a quantitative fashion in much larger data sets than would be possible using manual analysis approaches (see Figure 2.10-9).

Figure 2.10-9. Micro-CT of a mouse heart (left, radiopaque objects visualized from 5μm resolution data), identified coronary subtree in a maximum intensity projection image (middle, marked in white), and a 3-D representation used for quantitative analysis (right).

Quantitative Analysis of Coronary MDCT Images (Sonka, PI, 2003-2006, Philips Medical Systems). In direct collaboration with Philips Medical Systems, our group has developed a quantitative method for analysis of coronary morphology and plaque composition from 64-slice MDCT image data (2003-2006, Philips). The method will likely be used in the next generation of Philips CT analysis software.

Retinal OCT Image Analysis. We are developing a set of quantitative analysis tools for highly accurate assessment of retinal layers from clinically available optical coherence tomography (OCT) images. The retinal layers imaged by OCT provide information about the most important causes of blindness, such as glaucoma, macular degeneration, and optic neuropathy. Our current techniques allow simultaneous detection of multiple retinal layers and are being validated against functional parameters from the imaged patients. Figure 2.10-10 shows an example of retinal OCT segmentation and the resulting 3-D retinal thickness visualization.

• Retinal Color Photographs.

  Retinal color photographs are used for screening for causes of blindness, namely, macular degeneration and diabetic retinopathy. We have developed a comprehensive system for detecting lesions including hemorrhages, infarcts, and neovascularizations from these photographs. This system has been validated against retinal image analysis experts and is currently undergoing validation in a data set of 40,000 photographs.

• On-line Retinal Diagnosis (Abramoff, PI, 2005-2009, 1R01EY017066).

  A system for telediagnosis of retinal disease using retinal cameras in family care clinics across the Midwest is being built. This system is fully based on open-source components, allowing family physicians to diagnose patients with diabetes.

• Stereo-Photographs of the Optic Nerve.

  Development of algorithms and systems for automated classification of the optic nerve head from stereo photographs is being developed. So far, we have developed a method that outperformed all known algorithms for analyzing these noisy images. This system has been validated against six glaucoma experts.

• Functional Retinal Imaging.

  We are developing a method for measuring retinal responses to visual stimuli using infrared light sources. Collaborators have obtained excellent results in animal experiments and we are working to translate these results to humans.

Graph Search (Sonka, PI, 2006-2009, NIH-NIBIB - 6th percentile score, funding expected in June 2006). Medical image segmentation is of critical importance for a large portion of all quantitative medical imaging tasks. We have developed a novel method allowing optimal identification of 2-D, 3-D, and 4-D surfaces. This method has been recently extended to the analysis of multiple mutually interacting surfaces including cylindrical and spherical surfaces. The method is based on optimal graph-based image segmentation in which the optimality is assessed with respect to a task-specific cost function. We have already employed this approach to a variety of tasks and showed its utility in vascular image analysis using IVUS, MR, and ultrasound images, in bronchial tree analysis from X-ray CT, as well is in the OCT retinal images.

Image Registration (Christensen, Hoffman, PIs or PIs of subcontract, EB004126, CA096679, HL64368, HL073598)

• Non-Rigid Image Registration (NIR).

  NIR is an essential tool for morphologic comparisons in the presence of intra- and inter-individual anatomic variations. We are developing a framework for comprehensive NIR method evaluation that does not require a “Gold Standard” or ground truth correspondence map. The Non-rigid Image Registration Evaluation Project (NIREP) will extend the scope of prior validation projects by developing evaluation criteria and metrics using large image populations, richly annotated image databases, and computer-simulated data, and by increasing the number and types of evaluation criteria.

• Lung Trajectory Mapping for Intensity Modulated Radiation Therapy.

  The local control of lung cancer can be significantly improved, at fixed or reduced morbidity levels, by advances in 3-D imaging, treatment planning, and deliveries which overcome errors in planned and delivered dose due to breathing motion. Image registration is being applied to track lung motion using 4-D CT images with limited fields-of-view acquired over approximately 3 breathing cycles. The goal of the project is to successfully track the motion of the breathing lung using nonrigid image registration and to build a 4-D predictive model of the breathing lung.

• 3-D Imaging and Computer Model of the Respiratory Tract.

  This project is based at the Pacific Northwest National Laboratories. Our part is to adapt and apply image registration to build an atlas of the rat respiratory system similar to that of the human lung project described above.

• Image Registration Algorithm Development.

  We are continually developing new and improved image registration algorithms. One such recent algorithm is for surface-to-surface registration. Surface registration is apt for many medical imaging applications. Our new surface registration algorithm has several major advantages compared to other such techniques: it can jointly register a group of 3 instead of 2 images, and be extended to group-wise registration with more than 3 images in the group. It also provides better correspondence by minimizing inverse consistency and transitivity errors, which are difficult to minimize using other techniques.

Neuroimage Databases and Neuroimage-based Assessment of Clinical Trials in Psychiatry (Andreasen, Nopoulos, Paulsen, Moser, Robinson, Beglinger, Calarge, O’Leary, Block, Shoulson, 5T35-HL07485, R01-DE01-1439901, R01-MH031593, R01-NS640068, T32-MH019113, R01-MH40856, R01-MH60990, K23-AG020649, R03-AG024609, R01-MH63405, R01-MH65135, M01-RR0827, R01-DA010551, R01-DA019338, R01-HG02449).

A particular strength of Iowa’s Department of Psychiatry is its research using neuroimaging and related studies in image analysis and computer science. In fact, neuroimaging research on psychiatric subjects has been ongoing at Iowa for over two decades, beginning with CT studies and adapting to new technologies as they have developed. The first database-structured Magnetic Resonance (MR) study of schizophrenia was conducted at Iowa; subsequently, we have added Positron Emission Tomography (PET), Functional MR (fMR), and innovative image analysis strategies to our repertoire including MR Spectroscopy and MR Diffusion Tensor Imaging (DTI).

The foundation of our success in clinical neuroscience research has been the Schizophrenia Research Program directed by Nancy Andreasen, the largest and longest-running program in the nation for the study of schizophrenia. A core feature of this program is longitudinal neuroimaging, in which a total of nearly 400 patients have enrolled during the early phases of their illness, thereafter receiving repeat brain scans on a regular basis. This database of patients and healthy controls has been in operation for over 15 years.

The long-standing success of this research program has attracted a number of investigators to work in our department and trained others. Using some of the same tools, their work has expanded beyond the study of schizophrenia. Current neuroimaging projects include large-scale, multi-site projects, smaller local projects, and longitudinal projects whose subjects range from children to geriatric populations; the topics include the normal brain (gender differences, development, aging) as well as diseases and disorders. There is a high degree of integration among departments (neurology, pediatrics, radiology) and across disciplines (physicians, engineers, psychologists). All of these studies will utilize the infrastructure of the proposed Institute. Table 2.10-1 below provides examples of the types of studies currently underway as well as the scope of each project.

Table 2.10-1. Current Studies

Investigator	Type of Scan	# of Subject/year	Patient Type
Andreasen	PET/MRI	160	Schizophrenia
	MRI/fMRI	120	Schizophrenia
Beglinger	fMRI	45	Huntington's Disease
Block	PET/MRI	40	Marijuana users
	fMRI	24	Marijuana users
Calarge	MRI/DTI	20	Adults with Attention Deficit Disorder
	MRI/DTI	30	Children with Conduct Disorder
Ho	MRI	50	Siblings of patients with Schizophrenia
Jorge	fMRI	30	Vascular Depression / Post-Stroke
	Spectroscopy	20	Huntington’s Disease
Moser	PET/MRI	10	Atherosclerosis
	MRI/DTI	20	Atherosclerosis
Nopoulos	MRI	75	Children with Cleft lip/palate
	MRI/DTI	75	Children born prematurely
Paradiso	fMRI	20	Geriatric
Paulsen	MRI	550 from 24 sites	Huntington’s Disease
	fMR	30	Huntington’s Disease
O'Leary	PET/MRI	46	Marijuana users
Total		1365

Cardiovascular Clinical Trials (VanBeek, Weiner, Hoffman, 2005-2008, P30 CA086862, R01 HL064368). Cardiovascular imaging is increasingly making use of CT and MRI modalities for improved triage of patients. This has led to significant increases in chest imaging using CT angiography, both for the diagnosis of pulmonary vascular and coronary artery disorders. The introduction of improved CT scanners has increased the imaging speed as well as the spatial resolution of these procedures, resulting in high-quality images suitable for diagnostic management using only non-invasive methods.

The imaging-based approach to studying effects of interventions on diseases crucially depends on greater sensitivity of imaging methods. Novel systems (such as multidetector CT, multimodality imaging like PET-CT and MRI) and the introduction of computer algorithms for detection and quantification of diseases enhance this capability. Not only is this enhancement likely to benefit patients, but it is also projected to improve clinical trials by introducing alternative endpoints to study the effectiveness of new treatments, enabling earlier decisions on their introduction into the market. The economic benefits of such strategies are well understood by FDA, NIH and drug companies alike.

Medical Image File Archive and Retrieval (MIFAR) for Clinical Trial Support. At Iowa, we have built, tested, and integrated MIFAR into several of our clinical imaging studies - namely the Biomedical Research Partnership pulmonary project, the Muscatine cardiovascular risk factor study, National Emphysema Treatment Trial, etc.

Image-Based Surgery (Ryken, Howard, Industrial collaborations). The development of image guidance for neurosurgery has been a primary goal at The University of Iowa over the last 8 years. Dedicated efforts by Iowa’s Departments of Neurosurgery (and other surgical subspecialties including Otolaryngology and Orthopedics), Radiology and Radiation Oncology have resulted in the successful integration of three major clinical initiatives funded by hospital-based resources, creating a successful infrastructure for the delivery of image-guided subspecialty care. In 2005, over 300 image-guided procedures were performed for patients under the care of the Department of Neurosurgery alone, compared with 10 to 20 per year in the early 1990’s. This rapid expansion has provided new options for patient care and drives the continuous need for dedicated users of the technology to provide feedback for improvement in patient care and treatment delivery.

The current infrastructure is based on a centralized server, which acts as a repository for the accumulated multi-modal imaging data, post-processing (treatment planning) data and data collected from the treatment delivery. Developed initially as an alpha-study site with Surgical Navigation Technologies (now Medtronics Inc., Louisville, CO), this centralized-server-based platform remains one of the few systems of its kind and creates an ideal model for continued expansion and integration. Additional alpha and beta testing of technology has continued in a partnership effort, including advances in patient registration, integration of image-guided intraoperative ultrasound, integration of advanced volumetric image data sets and application of image-fusion and multi-modality image data sets. The platform is compatible with the treatment planning software used in the Department of Radiation Oncology, allowing streamlined integration for the care of patients requiring a combination of surgical and radiation clinical care.

Within the UIHC Department of Neurosurgery, the primary focus for additional development of this technology has fallen into two areas: improvement in the operating-room use of the technology and improvement in the treatment-planning processing of image data. For example, the incorporation of ultrasound into image-guided navigation in the operating room allows real-time feedback on intraoperative changes occurring during a surgical procedure while still allowing a standard image-guided technique to be utilized. The development of direct cranial fixation of the reference array required for image-guided procedures has allowed us to utilize image guidance during cranial procedures while keeping the patient awake for neurological monitoring. Incorporation of functional MR image data, computed positron emission tomography and magnetic resonance spectroscopy improves target selection and enhances the surgeons’ confidence in interpreting pre-operative imaging data for use during the procedure. Over the last 8 years, projects related to image guided neurosurgery have provided over 20 medical and biomedical students with training opportunities. Currently, funding for a collaborative effort with the Center for Non-Destructive Engineering at Iowa State University is allowing the further development of ultrasound for intra-operative treatment (HIFU – high-intensity focused ultrasound) as well as navigation, and the development of novel software tools for image enhancement and processing.

Image-Based Radiation Therapy (Buatti, Oberley, Domann, Spitz, 2003-2011, Center of Excellence in Image Guided Radiation Therapy, P01-CA66081, P01-CA66081, R01 CA073612, RO1CA112009). A critical barrier in clinical trials is the inability to accurately co-identify a biomarker, an intervention and subsequent effects. The center’s mission is to develop methods for integrating pre-intervention imaging that includes capacity for CT (4D-CT to account for respiratory movement), PET, 3T-MR including spectroscopy, and ultrasound. Thus, multi-modal, pre-treatment images of high resolution that are capable of providing functional information can be obtained and used for a given intervention. Software to fuse multiple sets of images is currently in place and used routinely. One potential intervention is stereotactically targeted radiation therapy that can be optically tracked in real time. Another potential capacity is stereotactically driven biopsy or tissue sampling prior to a localized intervention. The ability to identify a functionally relevant area anatomically (by virtue of the multi-modality imaging capacity) in stereotactic space will be critical in future trials aimed to assess the responses of tissues to interventions of many types. In addition, the capacity to sample and then resample at anatomically and potentially functionally consistent points will be critical to future trials assessing surrogate markers of response/success. Similarly, evaluation of tissue changes with high-resolution imaging over the course of therapies with the capacity for guidance of sampling or intervention will be crucial.

In the area of radiation therapy these imaging tools are used to create novel dose distributions that overlay multiple image sets including those accounting for respiratory motion and containing functional information. The planning is undertaken once this image processing is achieved and increasingly involves the development of algorithmic solutions for targeting sites to receive specific doses. For example, a head and neck region may include a tumor target to receive 70 Gy (as identified on PET), a secondary lymph node region at high risk for microscopic disease involvement (by CT), a portion of the normal parotid gland to receive less than 15 Gy and a spinal cord area (identified by MR or CT) to receive 40 Gy or less. Algorithms developed to limit critical areas to desired levels while uniformly avoiding other non-critical areas are an advance that is independent of image guided therapies. During or after a course of radiation therapy, a repeat image set may identify changes in both anatomic and functional information that could impact future therapy. Changes in PET uptake, perfusion, and spectroscopy are all candidates. If such hypotheses are tested, the capacity both to image and to sample tissues directly correlated to these changes is an essential core resource for future clinical translational research. Links to quantitative image analysis, a surgical intervention suite, and radiation delivery are all available within the Center of Excellence, which serves as a hub for translational research, with collaborations and joint appointments in engineering, nuclear medicine and surgical subspecialties.

Inter-University, International, and Industrial Collaborations. One of our areas of expertise is our ability to build inter-disciplinary and cross-disciplinary research teams which frequently include experts and/or end users from geographically distant locations. Over the years, we have gained a wealth of experience in inter-university research collaborations. For example, one of our NIH grants (Sonka, PI, 1999-2009, R01 HL 63373) includes cardiologists from the University of Chicago and Charles University in Prague, Czech Republic in addition to medical image analysis researchers and cardiologists from The University of Iowa. In another NIH-funded project (Sonka, PI, 2003-2007, R01 HL071809), University pediatric cardiologists, radiologists, geneticists, biostatisticians and medical image analysts collaborate with pediatric cardiologists from the Children’s Hospital in Boston.

Other projects include collaboration with a physiologist at the Mayo Clinic; with the Department of Radiology at Leiden University in the Netherlands; with the Graz Technical and the Medical Universities in Graz, Austria; and with Washington University in St. Louis. Similarly, the radiology group at Iowa participates in a Biomedical Research Partnership project (Hoffman, PI, 1999-2010, R01-HL64368) with researchers from Johns Hopkins University, the University of Washington, Auckland University Bioengineering Institute, and several other groups. In these collaborations, we frequently develop image analysis software tools to be used by non-engineers—physicians and technologists—in clinical and epidemiologic trials and for clinical care. Our software tool for quantitative analysis of endothelial function via measurement of flow-mediated dilatation in the brachial artery and measurement of carotid intima-media thickness is FDA-approved for clinical care and used in over 90 premier research institutions in North America, Europe, and Asia.

The University of Iowa has a comprehensive research agreement with Siemens Medical Systems, active research collaboration with Philips Medical Systems, with the midwestern company Vital Images, and licensing agreements with local medical image analysis companies VIDA Diagnostics and Medical Imaging Applications. All these avenues are used to achieve continuous dissemination of medical imaging and image analysis methods and software tools.