Research in Medical Imaging
Vision is the most important sense humans have. Computer vision tries to endow machines with similar capabilities to interpret visual input, and to act upon it. With the Medical Imaging team, we work on mainly four aspects:
- Virtual Reality in Medicine
- Medical Image Analysis
- Biomedical Simulations
- Biomedical Microsystems
To learn more about the individual topics and to show a list of related projects, please select a topic below:
Automatic Vein Pressure Measurement using Ultrasound
The goal is to develop an automatic system for the noninvasive measurement of peripheral venous pressure using ultrasound.
Venous pressure is an indication for several clinical situations and therefore its accurate measurement is an invaluable diagnostic tool. It is traditionally measured via catheters inserted into a subclavian or internal jugular vein. The former is known to yield unreliable results, and the latter is impractical, uncomfortable for the patient, and has potential complications.
The automatic ultrasound-based measurement system being developed involves an operator pressing gently on the forearm of a patient, while a computer processes ultrasound images along with pressure readings from a sensor placed between the ultrasound transducer and the skin. Venous pressure can then be estimated automatically from such measurements using image processing algorithms. This is a noninvasive system compared to the risks of catheter insertion based measurements. Furthermore, such an automatic system can yield faster, more accurate, repeatable, and observer-independent vein pressure measurement, compared to manual application of this method.
ReCoVa – Reconstruction of cerebrovascular networks
The general aim of the project is to reconstruct complete cortical cerebrovascular networks on the basis of partially incomplete, high-resolution tomographic data.
We are developing intelligent algorithms based on known morphological features of the cerebral vasculature. Ultimately we will generate a full vessel network based on insufficient data. Even if we focus on optimizing the results of an incomplete and error-prone measurement, the developed technology will in the future also support enhancing low resolution-images that cannot resolve smaller vessels, by filling up the missing parts of the vascular network based on a tissue footprint, either explicitly or in a statistical fashion. This would facilitate a reduction of imaging efforts substantially by moving to lower resolution imaging with the gain of a larger field of view e.g. of µCT instead of synchrotron radiation based tomography, or even extending the application domain to in vivo investigations relying on macroscopic imaging modalities like CT or MRI. Secondly, the methodology may later be used to combine individual patches of a sample to a larger volume by filling up the voids in a consistent fashion, even allowing the integration of image information extracted from different modalities.
Quantitative magnetic force microscopy (qMFM) is a technique to back out the magnetic structure of a sample from a measurement with MFM. The project objective is implement a reconstruction technique to evaluate the data and to package these tools in software that effectively hides complexity from the user.
In the very few cases where qMFM could be implemented so far, it provided otherwise inaccessible information on the magnetization mechanisms in thin films due to its high resolution, insensitivity to applied external fields, and calibrations validated independently of probe models. Here we seek to develop measurement techniques for reproducibly acquiring MFM data that can be evaluated quantitatively, to design numerical tools for this evaluation, and to package these tools in software that effectively hides complexity from the user.
We contribute algorithms for finding a correct and optimal calibration function for the implementation of qMFM, and performing the necessary image data deconvolutions. This involves taking into account priors such as e.g. knowledge of the negligible effect, far from the tip center, of high spatial frequency components. As a result of this work algorithms will be available to obtain optimal calibration functions from data obtained using highly accurate, automated measurement routines. Finally, our strategy for disseminating the knowledge obtained in this project is the establishment of web-based a framework for the effective utilization of these tools, including visualization capability and comprehensive tutorials.
Biomechanical Simulation of Transcatheter Aortic Valve Implantation
We investigate the biomechanical mechanisms behind Transcatheter Aortic Valve Implantation (TAVI) to reduce the risk assiciated with the intervention.
Transcatheter Aortic Valve Implantation (TAVI) has been established as the new standard treatment for high-risk patients with severe aortic valve stenosis. Thereby, a replacement valve consisting of three soft tissue leaflets fixed inside a metallic, foldable support frame (stent), is positioned and unfolded inside the diseased aortic valve under beating heart conditions. The native valve leaflets are dislocated by the unfolding stent and pressed against the vascular wall. The method is being applied worldwide, efficacy and safety have been proven superior to standard treatment of inoperable patients and at least non-inferior in patients with increased operative risk. Nevertheless, the technique is linked to a number of complications. Some are rather rare, but lethal, such as ruptures of vascular of ventricular tissue or the partial or complete obstruction of the coronary arteries by the native leaflets or by a misplaced stent. More frequent, postoperative arrhythmia or bradycardia as well as paravalvular leaks along the perimeter of the stented valve are witnessed. While not directly life-threatening, these complications will require additional correction (e.g. pacemaker) and are known to result in a worse long-term perspective. Our interdisciplinary research group at University Hospital Zürich and ETH Zürich is set out to investigate the biomechanical mechanisms behind these complications with the long-term aim to derive strategies for a patient-specific optimization of the planning as well as the conduction of TAVI.
University Hospital Zurich - Division of Cardiovascular Surgery (Prof. Falk)
FUSIMO - Patient specific modelling and simulation of focused ultrasound in moving organs
The aim of FUSIMO is to develop, implement and validate a multi-level model for moving abdominal organs for use in magnetic resonance-guided focused ultrasound surgery.
In recent years, High-Intensity Focused Ultrasound and Focused Ultrasound (FUS) have become frequent tools for non-invasive benign tumour therapy. However, treating tumors with focused ultrasound is challenging in terms of reliable therapy planning, monitoring and outcome prediction especially in moving organs with a complex blood supply. It is important to understand that the processes involved in FUS therapy are multi-level ranging from organ morphology, perfusion and motion, down to microscopic and cellular level. The relation within and between these levels is not well understood.
In this project, a multi-level model for moving abdominal organs for use in magnetic resonance-guided focused ultrasound surgery will be developed. The overall model will consist of several sub-models, which interact and describe aspects in a hierarchical manner. The integrated model will consist of:
- Abdominal organ model to simulate motion and the influence on ultrasound application
- Target organ/tumour model to capture organ/tumour physiology, and organ/tumour reaction to therapy
- Microscopic tissue model to simulate direct heat ablation, model energy distribution, tissue heating and cooling
- Model to evaluate first steps to simulate drug delivery, microbubble distribution and dynamics
Frauenhofer MEVIS, Germany,
University of Dundee, United Kingdom,
Technische Universiteit Delft, Netherlands
Stiftelsen SINTEF, Norway
Medical Imaging Research Institute, Germany
IBSmm Engineering spol. s r. o., Czech Republic
INSIGHTEC LTD, Israel
GEMS PET SYSTEMS AB, Sweden
UNIVERSITA DEGLI STUDI DI ROMA LA SAPIENZA, Italy
Eidgenössische Technische Hochschule Zürich, Switzerland
Fundatis Medis, Romania
An emerging trend in surgical simulation is the demand for custom systems by medical device manufacturers. This requires surgical simulator companies to provide highly specialized, tailored simulations, at low production numbers, in limited time.
Therefore, this project aims at creating a framework for rapid prototyping of customized systems. The starting point will be a generic simulator core. Various options to quickly extend the latter to specific needs will be addressed, including data-driven simulation methods.
VirtaMed AG, Switzerland
Eidgenössische Technische Hochschule Zürich, Switzerland
Weakly Supervised Segmentation for Building Statistical Models
The goal of this project is to investigate and develop unsupervised or minimally supervised techniques for the segmentation of the anatomy (in particular, bone structures).
Statistical models of variability in anatomical structures are essential for a wide range of medical applications, e.g., to reconstruct missing information in implant design or help localize bones for radiotherapy planning. Segmentation of the anatomy is a necessary step towards such robust statistical shape and intensity modeling. A major obstacle in generating such statistical models is the need for a sufficiently large amount of correctly segmented medical data, from which the statistical model can be derived. Besides the purposes of active shape modelling based segmentation, it is beneficial to develop prior-less or weakly-supervised techniques to prepare the initial data used for modelling, and to deal with cases not well described by the statistical shape model (e.g. trauma, pathology).
Conventional segmentation methods for such large sets on segmentation data require either a significant amount of manual, interactive effort of the medical personnel or a statistical shape model of the bone in question, and sometimes a combination of both. The objective of this project is to develop weakly supervised methods for segmentation for subsequent automatic generation of statistical models. To this end, we are utilizing single atlas methods to provide semantic context in a general segmentation framework. We are particularly interested in bone segmentation to be utilized in the generation of a statistical skeleton atlas of the body.
Flow Energy and Vessel Biology of the Cerebral Aneurysm
The goal of the proposed project is to continue developing and validating a numerical model that simulates the principal mechanisms involved in the formation and the endovascular treatment of aneurysms. Validation requires use of adequate methodologies to compare and match virtual with real world observations. Ex-vivo and in-vitro models, and where unavoidable, succinct use of animal models are needed to obtain highly valued biological validation information.
Energy of cerebral blood flow may lead in approx. 2-4% of people to vessel wall fatigue with creation of an intracranial aneurysm that may carry a risk of rupture. Vessel and blood biology are both critically influenced by shear, a measure of friction between moving elements of blood and the endothelial lining of the vessel wall. Clot formation is also closely involved in endovascular treatment with good and bad effects. Computational study of wall remodeling effects could include areas and quality of clot formation, that are understood to critically drive the biology of the vessel wall by inflammation and degradation mechanisms.As in any other field of biomechanical engineering, numerical simulations offer the possibilities to test a variety of scenarios and mechanisms allowing for forecasting disease evolution and for treatment planning. Since current medical imaging can provide exact 3D replicas of concerned patients, we wish to understand critical mechanisms that would help to predict rupture risk in unruptured aneurysms and in case of treatment need, to choose the optimal implant with the best chance to produce healing, i.e. reverse remodeling of the vessel wall. Once the interplay of clot formation and suitable endovascular implants is better understood, predictive techniques will support patientspecific, personalized assessment to increase the saftey of the intervention.
Image-based validation of computational models for vascularized solid tumour growth
In tumor modeling we focus on developing new approaches for a comprehensive validation of predictive models of tumor growth using image information.
While both, model development and tumor imaging, have seen rapid progress in recent years, rather little progress has been made so far towards assimilating experimental data into tumor models. We address the full integration of measurement and simulation by focusing on the initialization problem and the development of new methods for a fast image-based inference by combining generative models of disease progression with efficient techniques from machine learning. We rely primarily on functional imaging data from of small animal models for acquiring multimodal time series of tumor progression in vivo.
Evaluation and Validation of Ultrasound for the Management of Organ Motion in Tumour Therapy
Radiation therapy is the established method for non-resectable tumours. The art of radiation therapy is to deliver a lethal dose to all cancerous tissue whilst sparing as much healthy tissue as possible. Recent advances in three-dimensional (3D) planning and treatment technologies have enabled the delivery of highly conformal dose distributions. However exploiting the full potential of these highly localised treatment methods requires compensation for organ motion, which is substantial for tumours in the thorax and the abdomen. Beside quasiperiodic respiratory motion, the organ undergoes secondary modes of deformations caused for example by the cardiac cycle motion, digestive activity, gravity, muscle relaxation, or filling of the bladder. While methods have been proposed to compensate to some extent for perpetual breathing motion, they neglect the drift caused by the secondary modes and hence become inaccurate after a short period of time. In previous work, we have shown that the liver motion can be more accurately predicted by employing a statistical drift model updated by the current position of internal surrogate markers. Expanding on this theoretical analysis, we aim in this project at developing the techniques for real-time tumour location predictions based on the 3D position of one or more US tracked surrogate landmarks. Key milestones include efficient adaptive acquisition of 4D (3D+time) MRI data, syncronous acquisition of ultrasound and 4D MRI data, tracking of surrogate landmarks in ultrasound, prediction of organ motion based on surrogate marker position and statistical motion model, and evaluation of prediction accuracy. The 3-year project starts in January 2010 and is supported by the Swiss National Science Foundation.
Partners: Radiology Department, University Hosptial Geneva
Division of Radiological Physics, University Hospital Basel
Medical Image Analysis Centre, University of Basel
Centre for Proton Therapy, PSI
The target of this project is the development of a training simulator for knee arthroscopy.
Arthroscopy denotes the minimally invasive inspection and treatment of damage in joints under endoscopic guidance. This procedure demands specialized skills which need to be acquired by apprentice surgeons. One option for training is computer-based surgical simulation. This project focuses on the development of a nonhaptic version of a knee arthroscopy training simulator. A central goal will be to go beyond currently available systems by covering complete interventions, including large variability of patient cases, providing a realistic simulation interface (knee replica & instruments), and accurate modeling of the situs.
VirtaMed AG, Switzerland
Eidgenössische Technische Hochschule Zürich, Switzerland
University Hospital Balgrist, Switzerland
Zurich University of Applied Sciences, Switzerland
Modeling Tumor Growth and Angiogenesis
Objective: We are investigating carefully selected aspects of tumor growth and neovascularization, which is of major interest both for clinical care and basic biological research. Modeling the process of tumor development can not only improve our insight into the underlying mechanisms but will also contribute to the development of novel treatment strategies like optimizing dose delivery during radiation therapy or systematic selection of anti-angiogenic drugs depending on tumor phenotype.
Our goal is to develop a comprehensive simulation package, allowing to simulate the underlying physical and biochemical processes, covering mutual relationships between mechanical and biochemical stimuli, transport phenomena like flow, diffusive transport or active cell migration or possibly even gene expression, resulting in tissue formation, deformation or removal. The major aspects, which need to be accounted for, are listed below:
- Production, transport and absorption of
- molecules (nutrients, growth factors and inhibitors):(chemotaxis,chemokinesis)
- cells (haptotaxis,haptokinesis)
- Vessel growth (endothelial cell transport, growth, death)
- Tissue growth (tumor cell transport, growth, death)
- Blood flow in vessels
- Accurate biomechanical tissue model
Major emphasis will be given to the interaction and interrelation of these processes, in contrast to the currently existing approaches which usually only deal with one of these issues in an isolated fashion.
Partners: ETH Zürich, Animal Imaging Center
Cardiovascular diseases are one of the major causes of human morbidity and mortality, calling for novel and more efficient methods of treatment. Non-invasive diagnostic procedures are therefore of particular importance in managing the diseases. Among the different diagnostic modalities available today, magnetic resonance imaging (MRI) stands out as a potential single non-invasive tool for a comprehensive cardiac examination, allowing the assessment of vessel morphology and the quantification of blood flow through larger vessels. Despite its great potential, MRI exhibits limitations in particular when information with very high spatial and/or temporal resolution is needed. It has also been shown that inter-individual differences in vessel geometry have considerable impact on the flow regime. Accordingly, the applicability of generalized models and experimental setups for studying human blood flow is limited. Due to these restrictions it cannot be expected, that MR imaging can directly provide all data necessary for diagnosis and interventional planning. It has to be combined with advanced image analysis and simulation tools, which allow to extract and to reliably extrapolate data from the MR measurements, leading to a dense and detailed spatio- temporal description of the vascular flow. The objective of this project is therefore to create the fundaments of a comprehensive computational framework unifying appropriate MR imaging methods, image analysis, model building and visualization algorithms, and simulation techniques based on computational fluid dynamics (CFD), providing a highly efficient tool for the clinicians to select the best therapy for cardiovascular malfunctions.
Partners: Indian Institute of Technology, Kanpur, India
Acquisition and Modelling of Respiratory Organ Motion
Objective: The main goal of this project is to develop methods for the acquisition and modelling of respiratory organ motion and its intra-subject and inter-subject variability. Such realistic motion data are required to improve proton therapy planning in the presence of organ motion. For example, different dose delivery methodologies can be compared with respect to sensitivity to organ motion.
The developed 4DMRI method can capture breathing and its variability without using ionising radiation. Together with motion estimation by non-rigid registration, this permits for the first time to study the variability of respiratory motion over tens of minutes and provides valuable knowledge for numerous other treatments such as motion correction in radiofrequency (RF) tumor ablation or gating in MR guided Focused Ultrasound (MRgFUS).
Partners: Division of Radiation Medicine, PSI
Institute of Biomedical Engineering, ETHZ/USZ
Application of the Morel Atlas for target volume planning in functional neurosurgery
Performing functional neurosurgery without the possibility of direct neural stimulation requires the precise determination of the target position solely based on pre-operative MR data. This will be achieved by navigating the pilot MR images using the Morel histological atlas of the basal ganglia.
Currently only fixed geometrical relations between specific anatomical landmarks in appropriately selected datasets are used for the prediction of the stereotactical locations during interventional planning. The Morel atlas, however, implicitly contains much more information about the anatomical relations between the relevant structures of the basal ganglia.
In order to improve the precision of the interventional planning we will convert the currently existing atlas in a digital representation and develop statistical shape models to fully exploit its inherent potential.
We will then develop software tools to be integrated into the overall TcMRgFUS system for high precision interventional planning based on structural data gained from MR acquisitions. These tools will enable atlas-based target volume planning allowing to overlay the full set of stereotactic targets onto the MR image used for the planning process.