Semester and Master Projects at BIWI

We constantly offer interesting and challenging semester and master projects for motivated students at our lab. Below, you can find a list of topics that are currently being offered. Not all projects might be listed, if you are generally interested, do not hesitate to contact one of the supervisors, she/he might also give you an overview of other offered projects. Also, don't hesitate to contact us proposing your own ideas for projects, they are more than welcome.

In this presentation you can find an overview of our research topics and available projects.

Medical Applications

Before performing corrective operations for malunions (i.e. wrong bone healing) of the forearm meticulous planning is mandatory. Although angular deformities are easiliy detected, rotational malunions may pass unrecognoized and pose considerable difficulty in their exact assessment. Methods based on 2D image data are imprecise to obtain axial rotation of the radius bone. The aim of this work is to develop a reproducible method to assess the axial rotation of the radius bone in 3D models derived from MRI images of the forearm. For this project an interest in 3D modeling and image processing would be helpful. Programming skills in C++ are essential.

Supervisor(s):

Michael Emmersberger, ETZ J76.2, Tel.: +41 44 63 25318

PD Dr. Matthias Harders, ETF C107, Tel.: +41 44 63 25279

Robert Hudek, Balgrist Hospital, Tel.:

Professor:

Gábor Székely (szekely), ETF C117, Tel.: +41 44 63 25288

The target of this project is to explore automatic volumetric mesh generation for medical applications. Volumetric meshes are commonly employed in medical simulations as models for the tissue, such as in surgical planning. Conventional model generation schemes involves a segmentation step delineating the anatomy, followed by a meshing step generating elements conforming to this segmentation. This is often a cumbersome process, not only demanding the scarcely available time of health professionals, but also constituting a major bottleneck that prevents the automation of such modeling and overall simulation thereafter.

The focus of this project will be to investigate methods to accomplish such modeling. The interest will be on fast computational techniques, such that the developed method and its implementation can be used with complex anatomical scenes in a timely fashion.

The student will be provided by a main algorithmic framework. The tasks and the workflow will then be:
- to survey approaches in the literature;
- to develop the mesh generation framework;
- to investigate efficient computational architectures, e.g. GPU processing;
- to evaluate the developed method on medical images.

The extent of the project will be defined and finalized according to the interests and knowledge of the student. Please contact for more information and further details.

O. Goksel and S.E. Salcudean, "Image-Based Variational Meshing", IEEE Trans Medical Imaging 30(1):11-21, Jan 2011.

Supervisor(s):

Prof. Orçun Göksel, ETF C107, Tel.: +41 44 63 22529

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Surgical reconstruction of complex humerus (shoulder) fractures is one of the most challenging problems in orthopedic shoulder surgery. In order to facilitate the difficult surgical planning task, several computer-assisted methods were developed targeting the virtual reassembly of the fracture [1]. However, the accuracy of the virtual assembly primarily relies on the quality of the patient-specific 3-d models, acquired by segmenting the bone fragments from computed tomography (CT) data. As a consequence, the choice of the applied segmentation algorithm is critical for the overall planning outcome.

The goal of this project is to develop a method targeting the high-quality segmentation of fragments of fractured humeri. Since the segmentation of such type of data is challenging, the developed algorithm has to tackle several problems. Typically, a complex humerus fracture consists of multiple thin-walled bone fragments which are in close proximity and direct contact. Therefore, the algorithm must be capable of segmenting very thin fragments being in direct contact with each other without leakage. Additionally, the correct boundary between bone and soft tissue is hard to distinguish due to partial volume effects. The student can built up his work on several segmentation algorithms which were developed in [1]. Software development will be in C# or C++. The extent of the project will be defined and finalized according to the interests and knowledge of the student. Please feel free to contact for more information and further details.

This project is part of collaboration between the Computer Vision Laboratory of ETH and the University Hospital Balgrist. Therefore, you will have the possibility to have a working place in the recently formed computer-assisted research & development group (CARD) at the hospital. This this would be the perfect opportunity, if you are interested in gaining additional experience in the field of computer-assisted surgery.


Zur CARD Homepage

References: [1] DISS. ETH NO. 19102, Computer-Assisted Planning for Orthopedic Surgery, Philipp Fürnstahl, 2010

Supervisor(s):

Dr. (ETH) Philipp Fürnstahl, Univeristy Hospital Balgrist, Zurich, Tel.: 044 386 5746

Professor:

Gábor Székely (szekely), ETF C117, Tel.: +41 44 63 25288

The rotational movement of the forearm bones can be simplified expressed as a fixed-axis rotation of the radius around the ulna. In addition to the bone motion, the simulation of ligaments between the two bones (e.g. the interosseous membrane) is of great importance. The goal of this project is to develop a component of a surgical planning tool that simulates the deformation of user-defined ligaments based on the finite element method.

The project can be split into several parts (e.g. one for each student). First, the motion simulation will be implemented by using a single axis rotational model. Secondly, an already developed collision detection algorithm will be integrated for detecting bone impingements as well as collisions between ligaments and bones. The major part of the project is the actual ligament simulation using the finite element methods (master thesis only).

Software development will be in C# or C++ using the VTK library (experienced programmer preferred). Existing code (e.g. motion simulation, finite element) can be used in all stages of the project. The extent of the project will be defined according to number of participating students. Please feel free to contact for more information and further details.


Zur CARD Homepage

References: [1] DISS. ETH NO. 19102, Computer-Assisted Planning for Orthopedic Surgery, Philipp Fürnstahl, 2010

Supervisor(s):

Dr. (ETH) Fuernstahl Philipp, Uniklinik Balgrist, Zurich, Tel.: 044 386 5746

Professor:

Gábor Székely (szekely), ETF C117, Tel.: +41 44 63 25288

The movement of the knee joint consists of a coupled motion between the tibio-femoral and patella-femoral articulations. Since only little information on the kinematics of the knee is available, it is currently unclear whether knee motion must be considered for patient-specific planning of surgeries for knee joint diseases such as patellar pathologies. The goal of this project is to study and analyze the kinematics of the knee joint during flexion. Based on this work a simple patient-specific kinematic model should be developed which can be used for surgical planning (master thesis only).

The motion data to be analyzed is based on CT data acquired from volunteers. In a first step, 3-d models of all CTs have to be generated by applying existing segmentation methods. Thereafter, the relative motion of the knee bones (i.e. of the patella) during flexion will be analyzed. In a third step, a simple ligament model is developed based on user-defined insertion points of the ligaments and the motion data. In the master project, the major task is to additionally develop a simple but patient-specific kinematic model which can be integrated in the surgical planning. The model will include a basic (not physically correct) simulation of the ligaments.

Software development will be in C# or C++ using the VTK library (experienced programmer preferred). Existing code can be used in all stages of the project. The extent of the project will be defined according to number of participating students. Please feel free to contact for more information and further details.



Zur CARD Homepage

Supervisor(s):

Dr. (ETH) Fuernstahl Philipp, Uniklinik Balgrist, Zurich, Tel.: 044 386 5746

Professor:

Gábor Székely (szekely), ETF C117, Tel.: +41 44 63 25288

The aorta is the major artery leading away from the heart, and it distributes oxygenated blood to all parts of the body through the systemic circulation. The Coarctation of the aorta, abbreviated as CoA, is a congenital heart defect involving a narrowing of the aorta. CoA can cause high blood pressure or heart damage, and it accounts for almost 10% of congenital heart defects affecting tens of thousands of patients annually in the western world. To improve our understanding of the effect of a moderate CoA on the hemodynamics, the student needs to investigate the effect of both Newtonian and non-Newtonian rheological behavior on the hemodynamical characteristics, like the Wall Shear Stress (WSS) and the pressure gradient through the coarctation. Geometry segmentation and flow measurements computed from phase contrast magnetic resonance imaging (PC-MRI) will be disclosed during the project. To achieve this study, a substantial attention needs to be paid to the development of the numerical solver. The student needs to develop numerical tools to model blood flow as homogeneous fluid with non-Newtonian shear-thinning viscosity. Suitable algorithms, like Newton and damped Newton, need to be considered and implemented. Furthermore, the student needs to achieve a comparative study of some hemodynamical characteristics in the AoC and the healthy cases. Finally, the accuracy of the numerical computations should be demonstrated through numerical comparisons with both post-processed PC-MRI acquisitions and numerical results from the published literature. Several possible extensions are available for this project. In particular, we could consider the coupling with a two-phase mixture model that mimics the time evolution and the distribution of RBCs in the flow domain. Note: Please contact for more information and further details.

Supervisor(s):

Dr. Aymen Laadhari, ETF D114.2, Tel.: +41 44 63 26444

Professor:

Gábor Székely (szekely), ETF C117, Tel.: +41 44 63 25288

Breast cancer comprises 22.9% of all cancers in women. Survival rates can be greatly improved with early detection of a growing tumor, which can be screened routinely and hazard-free with medical ultrasound (US). However, many tumors show a low US echogenicity, which make their detection with conventional B-mode imaging challenging. The tumor tissue is stiffer than the healthy tissue, which increases the speed of sound and acoustic attenuation. We have developed and patented a hand-held ultrasound technology, which is able to reconstruct speed of sound and acoustic attenuation, providing accurate images of tumors in the female breast. In the next months, this technology will be evaluated for the first time in medical trials. Semester Thesis/Master thesis Topics: Breast ultrasound software for clinical tests with patients: Direct interface to research ultrasound machines (Analog Ultrasound). Incorporation of magnetic/optical sensors to track transducer position. Graphical User Interface (GUI) (QT/C/C++/Tablet PCs) optimized to minimize patient discomfort. Experimental validation with medical phantoms. Improvement of quality of ultrasound mammography: Survey of state-of-the-art beamforming, tracking, non-linear optimization and smoothness regularization methods. Optimization for real-time operation in Graphical Processing Units (GPUs) and High-Performance Computing Clusters (HPC). Tests on synthetic and experimental data. Three-dimensional simulation of medical ultrasound: Extension of available finite-difference time-domain software in Matlab® to High-Performance Computing Clusters (HPC) available in our lab, and to ETH central cluster resources (Brutus/Euler). Simulation of real ultrasound arrays and validation with experimental data.}

Supervisor(s):

Dr. Sergio Sanabria, ETF C110, Tel.: +41 44 63 24827

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Breast cancer comprises 22.9% of all cancers in women. Survival rates can be greatly improved with early detection of a growing tumor, which can be screened routinely and hazard-free with medical ultrasound (US). However, many tumors show a low US echogenicity, which make their detection with conventional B-mode imaging challenging. The tumor tissue is stiffer than the healthy tissue, which increases the speed of sound (SoS) and acoustic attenuation (AA). The computer vision laboratory at ETHZ developed and patented a hand-held ultrasound technology (https://www.ethz.ch/sparkaward), which is able to perform accurate tomographic reconstruction of SoS/AA, providing high-contrast tumor images. In the next months, this technology will be evaluated in medical trials. A major highlight of conventional US B-mode imaging is its capability to operate in real time, which allows for efficient screening of suspect tissue structures by smoothly adjusting the position and orientation of the US probe with respect to the body of the patient. However, the novel SoS/AA modalities require sophisticated software algorithms, which exceed the real time processing capabilities of conventional medical ultrasound systems. The acquired data is therefore currently processed off-line in a high-performance computing server. In order to optimize the screening procedure in medical trials, we now aim at incorporating portable computing platforms as an add-on to conventional ultrasound systems. This platforms will allow in-line processing of ultrasound SoS/AA data during medical screening. The topic of the offered semester/master thesis is the translation and optimization of signal processing from high-level programming languages (Matlab®, C/C++) into an FPGA architecture. The main focus is on advanced ultrasound beamforming and displacement tracking algorithms, which are essential building blocks of SoS/AA imaging. The algorithms will be implemented in a FPGA development board (Xilinx Kintex UltraScale), and benchmarked in terms of numerical accuracy and speedup achievement. Finally, a real time processing pipeline will be assembled with ultrasound raw data acquired with our medical ultrasound systems (Ultrasonix and Verasonics), and experimentally validated with medical phantoms. This project is also posted in the website of the Integrated Systems Laboratory of ETH Zurich (http://iis-projects.ee.ethz.ch/index.php/FPGA_acceleration_of_ultrasound_computed_tomography_for_in_vivo_tumor_screening)

Supervisor(s):

Dr. Sergio Sanabria, ETF C110, Tel.: +41 44 63 24827

Pascal Alexander Hager (IIS), ETZ J69.2, Tel.: (+41 44 63) 277 86

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Early detection of cardiac abnormalities is the first step in treating conditions that can be life-threatening. Computer-aided diagnosis tools can be fundamental for detecting abnormalities earlier than human eye can pick them up. Magnetic Resonance (MR) imaging is important for building such computational tools as it provides non-invasive in-vivo observations of the human heart. Most of the previous works focus on anatomical information for building computer-aided diagnosis tools. Novel acquisition techniques open up the opportunity to integrate functional information. In particular, time-resolved 3D MR velocity vector field mapping allows assessing and visualizing vascular hemodynamics non-invasively. The overall aim of this work is to investigate computational techniques and machine learning methods for diagnosing aortic valve stenosis and aortic dilation from time-resolved MR velocity field mapping of blood in the ascending aorta.

Majority of computer-aided diagnosis tools for detecting abnormal variations rely on image registration, i.e. aligning images onto a common reference frame so that corresponding pixels can be compared. The main difficulty here is that image registration is not particularly easy for vector-field images. The main objective of this project to develop computational tools that can perform automatic detection of the mentioned conditions with vector field mappings without explicit image registration. To this end, the project will first investigate automatic localization using vector field mappings and then develop classification techniques that are based on the initial automatic localization.

The project is a collaborative effort and will be co-supervised by Prof. Ender Konukoglu and Prof. Sebastian Kozerke. If you are interested in this Masters / Semester project please contact ender.konukoglu@vision.ee.ethz.ch.

Supervisor(s):

Prof. Ender Konukoglu, ETF E113, Tel.: +41 44 63 38816

Prof. Sebastian Kozerke, IBT ETHZ, Tel.:

Professor:

Ender Konukoglu (kender@vision.ee.ethz.ch), ETF E113, Tel.: +41 44 63 38816

Currently two concepts for intraoperative navigation are established in orthopedic surgery, namely surgical patient-specific guides and marker-based navigation systems. Surgical guides are custom-tailored and additively manufactured to fit the anatomy of a specific patient. Besides costs, a main drawback of the technique is the limited flexibility in the surgery. As the function of a guide must be predefined preoperatively, the surgeon has no possibility to adopt the navigation plan online. Marker-based navigation systems offer more intraoperative flexibility, but they have also technical drawbacks. Optical tracking requires a direct line of sight between the observed objects and the camera system. Moreover, the interface to the surgeon often entails standard input/output devices, which is cumbersome, and required dedicated personnel. The considerable drawbacks of both systems motivates research to develop new navigation techniques.

In this student project a software prototype should be developed, demonstrating the advantages of wearable augmented reality devices for the navigation of orthopedic surgeries. In a first task, a test bed has to be created based on plastic bone model, mimicking the situation observed in the surgery. The test bed will be used for the development of a registration algorithm, used to overlay the plastic bone models with a 3D preoperative plan. After the concept of pre- to intra-operative registration is established, the navigation support for one simple surgical task will be implemented by using the visualization and communication capabilities of the device to guide the surgeon through the task in a step-wise manner.

A student with strong software development skills is needed to complete the project successfully (preferable C#, unity). As theoretical background, the student should have experience in computer graphics and/or 3D reconstruction.

Supervisor(s):

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Fürnstahl Philipp, Balgrist University Hospital, Tel.: 044 510 7360

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Markdown to HTML

Introduction

The Finite Element Method (FEM) is generally used for simulation on soft bodies. FEM relies on a discretization of the physical domain, most often represented by a mesh. A suitable mesh is mandatory in order to get quantitatively and qualitatively correct results. However, a great amount of time and expertise is required to construct such a mesh, which generally consists of a large number of elements, making computations computationally expensive. Muscles are soft bodies with nonlinear material properties displaying smooth, yet non-trivial shapes and possess a main direction following the muscle fibres. Modelling muscles with B-spline volumes has several advantages over the classic mesh representation: (1) significantly fewer elements are needed to preserve the smoothness and accuracy; (2) elements parametric coordinate frame can be aligned to efficiently represent muscle fibres; (3) the modelling can potentially be performed with much less manual user input.

Task Description

In the main part of this project, you will extend an existing codebase, in order to generate volumetric models of muscles with appropriate parametrizations. The second part of the work consists in performing the simulation of the models and their comparison to more standard meshes. The latter will be carried out in the SOFA framework (sofa-framework.org).

Skills

  • Solid programming skills in C++
  • Motivated student
  • Background or interest in 3D modelling and simulation

Remarks

The thesis is a joint collaboration between CAiM (caim.ee.ethz.ch) and CGL (graphics.ethz.ch). A written report and an oral presentation conclude the thesis. The thesis will be overseen by Prof. Gross and Prof. Göksel, and supervised by Dr. Tobias Martin (ETH, CGL) and Fabien Péan (ETH, CAiM).

Supervisor(s):

Fabien Péan, ETF C111, Tel.: +41 44 63 27632

Martin, Tobias, , Tel.:

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

A Computer Assisted Surgery Planning Application (CASPA) has been developed by the CARD team as a software platform for the assisted 3D planning of orthopedic surgeries. One of the functionalities of CASPA, is the surgical planning of osteotomies (anatomical reduction of bones), which is used to achieve the restauration of a normal bone anatomy. Solutions obtained for the surgical planning are subject to a large range of parameters and the quality of the solution is dependent on the achievement of specific clinical goals and optimization targets. Moreover, the surgeons’ expertise and choice preferences play a crucial role on the refinement of the final output of the planning.

Goal of this student project is the development of a decision maker interface to improve the set of solutions generated from an existing optimization algorithm. The expected decision maker interface comprehends the following milestones:

(1) Implementation of a clustering algorithm to group similar solutions according to the different optimization targets.

(2) Development of a graphical user interface for presenting the clustered solutions to the user. The GUI should allow the user to select a defined number of solutions and provide inputs about parameter range and importance of the optimization targets. Good visualization techniques are required for providing intuitive and easy-to-use controls and UI elements.

(3) Adaptation of a state-of-the-art decision maker (DM) approach [1-4], using the information from user input and existing solutions. The DM should narrow down the parameter range and provide a weighting of the different optimization targets for a second round of the optimization algorithm.

(4) Evaluation of the decision-making interface using the existing optimization algorithm. The interface should allow the generation of improved solutions of the planning, according to the choices made by the user.

The project builds upon existing optimization algorithms, and also on rendering techniques already included in CASPA. The main focus of the project is on the usability interface. The complexity of the DM can be agreed on according to the student background and the duration of the thesis.

The project will be jointly supervised by the Computer Assisted Research Development (CARD) group at the University Hospital Balgrist Zürich and the Interactive Graphics and the Computer Vision Lab (CVL) of ETH. You should have a solid background in C# / C++ and Graphical User Interfaces. Knowledge in machine learning is advantageous but not required.

You will have the opportunity to work in a practical and interdisciplinary environment, and to be directly involved on the improvement of the pipeline of pre-operative planning for orthopedic surgeries.

References:

[1] M. Tanaka, H. Watanabe, Y. Furukawa, and T. Tanino, "GA-based decision support system for multicriteria optimization," in Systems, Man and Cybernetics, 1995. Intelligent Systems for the 21st Century., IEEE International Conference on, 1995, pp. 1556-1561.

[2] C. Stephanidis, C. Karagiannidis, and A. Koumpis, "Decision making in intelligent user interfaces," in Proceedings of the 2nd international conference on Intelligent user interfaces, 1997, pp. 195-202.

[3] K. Deb and A. Kumar, "Interactive evolutionary multi-objective optimization and decision-making using reference direction method," in Proceedings of the 9th annual conference on Genetic and evolutionary computation, 2007, pp. 781-788.

[4] J. Branke, K. Deb, H. Dierolf, and M. Osswald, "Finding Knees in Multi-objective Optimization," in Parallel Problem Solving from Nature - PPSN VIII: 8th International Conference, Birmingham, UK, September 18-22, 2004. Proceedings, X. Yao, E. K. Burke, J. A. Lozano, J. Smith, J. J. Merelo-Guervós, J. A. Bullinaria, et al., Eds., ed Berlin, Heidelberg: Springer Berlin Heidelberg, 2004, pp. 722-731. CARD Homepage

Supervisor(s):

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Fabio Carrillo, CARD Group, University Hospital Balgrist, Tel.: +41443865744

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Recent advances in ultrasound simulation allow for simulating realistic ultrasound images in interactive times. This makes it possible for doctors to train certain difficult medical procedures in Virtual Reality simulations before applying them on actual patients, hence raising the success rate of these procedures. Recent US simulation models use a combination of surface-based ray tracing and volumetric ray marching, and also utilize the capabilities of modern GPUs (Graphics Processing Units) for parallel ray evaluation, e.g., by using the high-level Optix GPU ray tracing library. In this project, the goal is to develop a low-level CUDA based ray tracing application for ultrasound, replacing an earlier Optix-based ray tracing approach. This will give us maximal control over scheduling and casting of the rays, allowing us to investigate schemes for optimal parallelization and hence optimal throughput. Accordingly, a highly optimized US simulation will be developed, increasing the sophistication and hence realism at interactive times. We welcome students that are interested in medical ultrasound image formation, ray-tracing, and GPU programming.

Supervisor(s):

Dr. Oliver Mattausch, ETF C106, Tel.: +41 44 63 20714

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Ultrasound elastography is a method that allows the noninvasive measurement of mechanical properties (relative stiffness, Young’s modulus, Bulk modulus, viscosity, nonlinearity) of in vivo soft tissue with emerging clinical applications from breast tissue characterization to cardiology. Harmonic elastography involves applying multiple frequency steady state vibrations to tissues in order to measure the aforementioned properties. In order to evaluate our methods, tissue mimicking phantoms (gelatin, silicon) with different properties are built and tested on. It is the aim of this project to create a simple rheometry setup for measuring the mechanical properties of different tissue mimicking phantoms. The setups can be as simple as placing the phantom on a scale, applying a weight on top of it and measuring the axial deformation to determine its Young’s modulus, or slightly more complex. Phantoms with different properties such as materials (gelatin, agar, and silicon), sizes, stiffness as well as different inclusions sizes, stiffness and position within the phantom will be created and tested in the rheometry setup. Furthermore, the effect of different scatters (flour, cellulose) will also be tested (e.g. signal damping, signal traceability). In the case of a master thesis, experimental results will be compared with finite element simulation results from ANSYS. Type of work: Semester Project or Master Thesis, 40% Theory, 25% Software implementation, 35% Experiments. Requirements: Mechanical design principles, basic Matlab / C++ knowledge, Ansys knowledge is a plus, but not necessary.

Supervisor(s):

Corin Felix Otesteanu, ETF D112, Tel.: +41 44 63 38815

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Congenital deformities or wrong post-traumatic healing positions of the forearm bones can cause limitations in the range of motion (ROM) of patients, generate pain, engender instability of the joint and cause aesthetics problems. Anatomical reduction of the affected bones by surgical treatment (osteotomy) is used to achieve the restauration of a normal joint/bone anatomy. Cutting-edge technologies for osteotomy procedures include 3D pre-operative planning of the surgery, virtually performed on patient-specific bone models generated through image segmentation of medical CT. However, one of the current challenges in preoperative planning is the omission of surrounding muscles, articulations and further soft tissue structures. Inclusion of accurate patient-specific soft tissue information is one of the on-going research areas.

Goal of this student project is the development of a biomechanical model of the forearm. The project builds upon available 3D models of forearm bones and existing GUI for operation planning, and includes:

(1) Literature research for state-of-the-art in atlas fitting and soft tissue simulation of the forearm, focusing on the interosseous membrane (IOM).

(2) Selection of an (existing) anatomical atlas of the forearm and implementation of deformable registration of the atlas to the patient specific bone data, using similar techniques as presented in [1, 2].

(3) Development of a biomechanical model /simulation of the IOM of the forearm. This part constitutes the main part of the project. The idea is to be able to simulate/analyze the motion of the forearm with the patient specific data, and adapt the previously fitted atlas, focusing on the behavior of the IOM [3-5].

(4) Evaluation and validation of the model through available clinical pre and post-operative data.

The project will be jointly supervised by the Computer Vision Lab (CVL) of ETH and the Computer Assisted Research Development (CARD) group at the University Hospital Balgrist. Objectives and scope could be adapted according to the duration of the project (master/bachelor or semester thesis).You should have a solid background in computer graphics and be familiar with 3D modelling. Experience in human anatomy and biomechanics are advantageous.

References:
[1]B. Amberg, S. Romdhani, and T. Vetter, "Optimal step nonrigid icp algorithms for surface registration," in Computer Vision and Pattern Recognition, 2007. CVPR'07. IEEE Conference on, 2007, pp. 1-8.
[2] H. Seim, H. Lamecker, M. Heller, and S. Zachow, "Segmentation of bony structures with ligament attachment sites," in Bildverarbeitung für die Medizin 2008, ed: Springer, 2008, pp. 207-211.
[3] T. Heimann, F. Chung, H. Lamecker, and H. Delingette, "Subject-specific ligament models: Toward real-time simulation of the knee joint," in Computational Biomechanics for Medicine, ed: Springer, 2010, pp. 107-119.
[4] H. J. Pfaeffle, K. J. Fischer, A. Srinivasa, T. Manson, S. L. Y. Woo, and M. Tomaino, "A Model of Stress and Strain in the Interosseous Ligament of the Forearm Based on Fiber Network Theory," Journal of Biomechanical Engineering, vol. 128, pp. 725-732, 2006.
[5] J. Pfaeffle, J. Weiss, J. Gardiner, K. Fischer, T. Manson, M. Tomaino, et al., "The stress and strain distribution in the interosseous ligament of the human forearm varies with forearm rotation," in Trans 46th Meeting Orthop Res Soc, 2000, p. 140.

CARD Homepage

Supervisor(s):

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Fabien Péan, ETF C111, Tel.: +41 44 63 27632

Fabio Carrillo, CARD Group, University Hospital Balgrist, Tel.: +41443865744

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Mechanical exciters for ultrasound harmonic elastography Status: Still vacant Background Ultrasound elastography is a method that allows the noninvasive measurement of mechanical properties (relative stiffness, Young’s modulus, Bulk modulus, viscosity, nonlinearity) of in vivo soft tissue with emerging clinical applications from breast tissue characterization to cardiology. Harmonic elastography involves applying multiple frequency steady state vibrations to tissues in order to measure the aforementioned properties. Project aim It is the purpose of this project to work with mechanical exciters in order to generate such multi frequency vibrations. The exciter will be controlled and programmed on a PC via Matlab or C++. Synchronization between imaging and excitation will also be developed. The performance of our current exciter (BEI Kimco LAS-16 VCA) will be compared to those of a regular loudspeaker/ loudspeaker array, experimentally, on gelatin phantoms. Type of work Semester Project of Master Thesis 35% Theory, 35% Software implementation, 30% Experiments Requirements Skills in Matlab / C++ programming Supervisor Corin Otesteanu, corino@vision.ee.ethz.ch, ETF E13, phone: + 41 44 63 38815 Professor Orcun Goksel, ogoksel@vision.ee.ethz.ch, ETF C107

Supervisor(s):

Corin Felix Otesteanu, ETF D112, Tel.: +41 44 63 38815

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

MASTER / SEMESTER PROJECT

Predicting spinal surgery outcome for the anatomy of trunk

 

Deformities of the spine caused by an accident or by a birth defect may cause severe pain and functional impairment. Spine reconstruction by surgical treatment is used to achieve spinal balance, stability, and anatomical alignment. One important aspect for patients who undergo spinal surgery is the aesthetic aspect of the surgical outcome. It would be very helpful to be able to predict the postoperative trunk appearance of a patient, already in the preoperative planning stage.

A first step in simulating postoperative appearance is to predict the appearance of the surrounding anatomy (e.g., shape, position) after surgery which will be the goal of this project. That is, given triangular surface models of the trunk anatomy (e.g., spine, other bones, soft tissue structures) before surgery and the surgical plan (i.e., the modification of the spine) , a method has to be developed that simulates the effects on the surrounding anatomy. In the project, it is not necessary to achieve a biomechanically realistic simulation, visual appearance may be able to predict with other lower-complexity simulation techniques.

The project will be jointly supervised by the Computer Vision Laboratory of ETH and the University Hospital Balgrist. Required fundamentals in anatomy and clinical biomechanics will be provided by a dedicated supervisor at Balgrist.

You should have a solid background in computer graphics and/or simulation and strong programming skills. Experience in VTK and SOFA are advantageous.

Supervisor(s):

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Fabien Péan, ETF C111, Tel.: +41 44 63 27632

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

MASTER / SEMESTER PROJECT

Topographic reconstruction of legs for postoperative appearance after realignment surgeries

 Bone deformities of the leg (“O-Bein”, “X-Bein”) result in a deviation of the mechanical leg axis which may cause disproportional loading in the knee joint. Ultimately, osteoarthritis can develop and joint replacement by a prosthetic implant might be necessary. A surgical procedure (so-called osteotomy) can be performed to correct the leg axis by cutting and realigning the bone according to a preoperative plan. Besides restoration of the joint function, aesthetic aspects are also important for the patient because realignment of the leg axis positively influences the visual appearance. Therefore, the simulation of the postoperative appearance emphasizing photo-realistic visualization is required.

Goal of this student project is to develop a simulation pipeline using preoperative data (photographs, X-ray images) that can be used to predict the surgical outcome by rendering a photo-realistic visualization of the legs. The project builds up on existing methods and equipment, but also requires the development of new strategies optimized for the given scenario of patient image acquisition and model generation. In a first step, the workflow for generating a 3D textured appearance model from the patient leg before surgery will be developed. Second, the appearance model has to be transformed according to the surgery to be performed.

The project will be jointly supervised by the Computer Vision Laboratory of ETH and the University Hospital Balgrist. Required fundamentals in anatomy and clinical biomechanics will be provided by a dedicated supervisor at Balgrist. You should have a solid background in computer vision and strong programming skills. Experience in 3D model reconstruction and related standard methods (e.g., VisualSFM) is advantageous.

 

Supervisor(s):

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Statistical Shape Modeling is a valuable tool in image analysis. Correspondence establishment is a critical step in such shape analysis. It allows for tracking points across different instances of the same shape, and thus analysing the variability of such points within the population. The resulting statistical shape models are a very important tool for several clinical applications, from population studies to implant design.

The goal of the project is to investigate methods for establishing such correspondences between surfaces in a population of images. Common challenges in this step are the surface meshing and correspondence establishment in ambiguous or convoluted anatomical surfaces. In order to overcome these challenges, methods which explore the consistency in correspondences among multiple shapes are necessary.

The student will survey similar approaches from the literature; implement ideas using Matlab and C++ and evaluate results comperatively on CT and MR image datasets. Please contact for more information and further details of this project.

T. Gass, G. Szekely, O. Goksel, "Detection and correction of inconsistency-based errors in non-rigid registration", In SPIE Medical Imaging, pp. 8, no. 9034-46, San Diego, CA, USA, Feb 2014.

Supervisor(s):

Firat Özdemir, ETF C111, Tel.: +41 44 63 27685

Prof. Orçun Göksel, ETF C107, Tel.: +41 44 63 22529

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529



Minimal invasive tumour therapies are getting ever more sophisticated with novel treatment approaches and new devices allowing for improved targeting precision. Applying these effectively requires precise localization of the structures of interest. Respiration induces organ motion which cannot be neglected during therapy. Motion Models have been developed for predicting from partial observations the organ position.

Creation of such population models from 4D-MRIs and individualization of these models to a specific subject require many sub-tasks, including intra-subject registration of the 4D-MRIs, segmentation of structures of interest and definition of inter-subject correspondences. The last two tasks have so far mainly been based on manual interactions to avoid delaying the development of the abdominal motion models. However this makes the creation of the models quite labour intensive, especially as the number of subjects grows and more structures (liver, ribs, spine, kidney etc.) shall be considered. Therefore a method to (semi-) automate one of these sub-tasks shall be developed in this Master project.

Tasks
- Understand manual segmentation methods for creating motion model and available ground truth data
- Formulate approach for segmentation method, which exploits ground truth annotations
- Implement segmentation method
- Assess performance of method in comparison to ground truth annotations and with respect to overall motion model performance

Supervisor(s):

Dr. Christine Tanner, ETF C108, Tel.: +41 44 63 26246

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Generation of 3-d models based on medical imaging is a fundamental and particularly important step for the diagnosis of diseases on the one hand as well as for patient specific computer assisted planning applications on the other hand. Anatomical reconstruction of complex pathologies relies on an optimal visualization of the underlying conditions. Hence an accurate segmentation algorithm is essential for the success of subsequent planning tasks (i.e. fracture reconstruction [1], joint replacement surgery), including the comparison and registration of parts of the skeleton.

The goal of this project is to develop a method for (semi-)automatic high-quality segmentation of the upper arm, including the shoulder and the elbow. As the density of long bones varies along the longitudinal axis and bone is less dense near the joints, the algorithm has to take the local distribution into account to avoid leakage. Additionally, the correct boundary of the bones in cases of joint space narrowing (less distance between the articulating parts) has to be distinguished.

The student will gain immediate assistance by the technical experts, software engineers as well as for the medical part by an orthopedic surgeon, working at this field. Basic knowledge in Computer Vision and intermediate programming experience is required, however no medical background is necessary. Software development should be in C# or C++ and will be based on the already developed planning software CASPA and subsequent implemented as a tool. The extent of the project will be defined and finalized according to the interests and knowledge of the student. Please feel free to contact for more information and further details.

CARD Homepage:
www.card.balgrist.ch

Second Supervisor:
Dr. Philipp Fürnstahl, PhD, Head of CARD Team, University Hospital Balgrist, Zurich
Phone: +41 44 386 5746
Email: philipp.fuernstahl@card.balgrist.ch

References:
[1] DISS. ETH NO. 19102, Computer-Assisted Planning for Orthopedic Surgery, Philipp Fürnstahl, 2010

Supervisor(s):

Firat Özdemir, ETF C111, Tel.: +41 44 63 27685

Dr. Philipp Fuernstahl, ETF D115, Tel.: +41 44 63 27731

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Abdominal motion due to breathing is a major source of error during radiation therapy for cancer treatment. In order to establish techniques to correct such errors, those errors should first need to be tracable. Registration is a common tool in image analysis to estimate dense motion fields. Despite techniques to account for sliding inter-organ abdominal motion [1], registration of individual pair-wise images are prone to errors.

This project will invesitgate the use of information from the consistency of several individual registrations [2], in order to improve the accuracy of breathing motion recovery from temporal Magnetic Resonance (MR) images. The project will be conducted in Matlab or C++, depending on the expertise of the student. Please contact for more information and details on this project.

[1] Valeriy Vishnevskiy, Tobias Gass, Gabor Szekely, and Orcun Goksel, "Total Variation Regularization of Displacements in Parametric Image Registration", In MICCAI Workshop on Abdominal Imaging: Computational and Clinical Applications, Boston, MA, USA, Sep 2014.
[2] T. Gass, G. Szekely, O. Goksel, "Detection and correction of inconsistency-based errors in non-rigid registration", In SPIE Medical Imaging, pp. 8, no. 9034-46, San Diego, CA, USA, Feb 2014.

Supervisor(s):

Dr. Christine Tanner, ETF C108, Tel.: +41 44 63 26246

Prof. Orçun Göksel, ETF C107, Tel.: +41 44 63 22529

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Markov Random Feilds is a generic solution scheme, successfully applied on image segmentation tasks. Often the main challenge in medical image segmentation is the incorporation of user interaction. Although there exist several automatic segmentation methods in the literature, including the ones devloped in our group; it is nevertheless not uncommon that these may periodically fail due to imaging artifacts (e.g. implants that the patient has), type of image preprocessing, and complex anatomical regions of interest. The remedy for such situations is typically for a medical expert to then manually segment that failed case, which is a laborous task considering the scarcely available time of health professionals.

The goal of this project is to develop tools to assist the medical expert in their segmentation by efficiently incorporating computer vision techniques into the workflow. Experience in Matlab, C++, or Phyton will be among necessary skills. This efficacy of the methods will be evaluated by the student using the available image datasets.

The extent of the project will be defined and finalized according to the interests and knowledge of the student. Please contact for more information and further details.

T. Gass, G. Szekely, O. Goksel, "Simultaneous Segmentation and Multi-Resolution Nonrigid Atlas Registration", IEEE Trans Image Processing 23(7):2931-43, July 2014.

Supervisor(s):

Prof. Orçun Göksel, ETF C107, Tel.: +41 44 63 22529

Firat Özdemir, ETF C111, Tel.: +41 44 63 27685

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Intracranial aneurysms are malformations of cerebral arteries that are typically saclike in shape and occur close to vessel bifurcations, see Fig 1. Aneurysms indicate a region on the arterial wall with an increased risk of rupture, which is a dangerous event for the patient and may lead to permanent brain damage or death. Intracranial aneurysms are prevalent, roughly 3-5% of the population is affected, but the annual rupture risk of an aneurysm is below 1% in average, which makes the clinical decision-making (whether to treat an aneurysm or not) a difficult task. As they normally do not cause any symptoms, unruptured intracranial aneurysms are diagnosed incidentally. They can be identified in so-called angiographies where the brain vasculature is contrasted from other brain tissues. If clinicians were equipped with a tool that helps them to assess the status of an aneurysm or to match its morphological characteristics with those registered in a database, clinical risk assessment could be considerably improved. This semester or master project will be about the robust segmentation and/or isolation of intracranial aneurysms for different imaging modalities (CTA, MRA and 3DRA) and how their shape characteristics depend on aspects like resolution, contrast or the presence of imaging artifacts. The detailed scope of the project can be matched with the interests and expertise of the candidate at the beginning of the project. A good understanding of computer vision and coding skills are mandatory however. Ambitious students will get the possibility to work on a publication. A rich set of clinical imaging data is already available. The thesis will contribute to the AneuX project and will be conducted in collaboration with clinicians at the University Hospital of Geneva and the Hirslanden Clinic in Zurich. So the student will be offered the chance to get insights into clinical practice.

Supervisor(s):

Prof. Dr. Sven Hirsch, extern, Tel.: +41 44 63 23963

Norman Juchler, Institute of Physiology, University of Zurich, Tel.:

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529

Motivation

Optical coherence tomography (OCT) is an optical imaging technique which measures backscattered light to provide high resolution (1-10µm) cross-sectional images of soft biological tissue. In vitreoretinal surgery, an OCT-probe can be connected to a surgical instrument to provide information about the distance to, and the thickness of the retina. Extracting this information from a 1 dimensional image (A-scan) has proven quite challenging. A first test with a simulated set of annotated A-scans has shown that Deep Learning techniques (a Convolutional Neural Network in this case) can be used to detect the retina in these A-scans. However, this is a preliminary test, and it is not yet possible to recognize the thickness of the retina or to find retinal layers or blood vessels.

Goal

The goal of this project is to develop a neural network that can detect and track landmarks in OCT A-scans (see Figure), e.g. the retina. You will also research the possibilities of using Recurring Neural Networks, to utilize a-priori knowledge about the movement of probe and the environment.

Supervisor(s):

Kevis-Kokitsi Maninis, ETF D115, Tel.: +41 44 63 39063

Dr. Jordi Pont-Tuset, ETF D114.2, Tel.: +41 44 63 26667

Professor:

Luc Van Gool (vangool), ETF C117, Tel.: +41 44 63 26578

In this Master project the student will investigate novel semi-supervised machine learning techniques for the automatic delineation (also known as segmentation) of structures in MR images of the heart. Specifically, the project will be focused on new ideas from the field of neural networks and deep learning.

Description

Deep learning methods are currently revolutionising the fields of computer vision and medical image analysis alike. Given enough training data almost any prediction task can be learned to a very high degree of accuracy. However, particularly in the domain of medical images, obtaining large quantities of labelled training data can be very costly. For instance, it takes an experienced doctor multiple hours to delineate (i.e. segment) the compartments of the heart in a 3D MR image. Moreover, often these tasks can only be performed by medically trained personnel.

In the case of the heart such segmentations can provide important clinical information of the normality/abnormality of the cardiac function and form the starting point for population studies which define normality in the first place. While labelled data is hard to obtain, at the same time, every hospital has thousands of unlabelled data stored away for future reference which is currently not used.

Semi-supervised machine learning techniques attempt to combine labelled data with potentially large sources of unlabelled data to improve the accuracy of a learning task. Very recently, a number of approaches have been proposed which make use of new developments in deep learning such as autoencoders and generative adversarial networks. However, there is little to no work on semi-supervised learning for segmentation and specifically for segmentation of medical images.

In the this project, the student will:

  • investigate and extend the state-of-the-art in semi-supervised learning to cardiac segmentation and related applications
  • implement his or her methods in a modern deep learning framework such as tensorflow

This is a challenging project and the student should ideally have a strong background in programming and maths, in particular in probability theory. The project will be divided roughly into 50% development, 40% formulating new ideas/literature review and 10% writing of the thesis. In case of interest by the student, there is a good chance that this work may result in a publication.

The location of the project will be in the D-ITET computer vision lab (CVL) which is located in ETF in ETH Zentrum.

Goals

  • Demonstrate the feasibility of semi-supervised learning for medical image segmentation
  • Implement techniques in modern deep learning framework such as tensorflow
  • Thoroughly evaluate technique for thesis and potentially for a publication

Supervisor(s):

Dr. Christian Baumgartner, ETF E112, Tel.: +41 44 63 24893

Professor:

Ender Konukoglu (kender@vision.ee.ethz.ch), ETF E113, Tel.: +41 44 63 38816

Congenital fetal malformations are often associated with a poor prognosis and in many cases result in a severe socioeconomic burden. Recent advances in sequence and hardware technology enabled fetal magnetic resonance imaging (MRI) to become a clinically viable diagnostic modality for many indications. Our project builds on in utero fetal MRI using two, emerging acquisition techniques: diffusion-weighted and diffusion tensor imaging of the fetal brain (DWI) and intravoxel incoherent motion imaging of the fetal organs and the placenta (IVIM). DWI and IVIM are confounded severely by fetal movements and maternal breathing. The candidate will develop and implement advanced image processing approaches that will analyse fetal movements and correct for organ movements. This would result in the better estimation of diffusion- and IVIM parameters, and will mean better clinical diagnosis for selected congenital somatic (such as lung, heart malformations or fetal tumors) or brain abnormalities.

Tasks:

- Implementing a medical image processing pipeline that corrects for the motion of fetal organs on in utero acquired, diffusion-weighted magnetic resonance images

- Specifically, to implement a non-linear deformation, optical flow or machine learning algorithm to track the location of moving fetal organs in DWI and/or IVIM images with high background noise

- To publish the work at a relevant scientific conference or in a journal publication

Requirements:

Programming knowledge (e.g. python, Matlab or shell scripting and knowledge in using medical image processing tools)

- Experience in image processing, especially (non-linear) image registration, object tracking algorithms

- Good English knowledge

Literature:

Jakab A, et al. In utero diffusion tensor imaging of the fetal brain: A reproducibility study. Neuroimage Clin. 15:601-612.

Moore RJ, et al. In vivo intravoxel incoherent motion measurements in the human placenta using echo-planar imaging at 0.5 T. Magn Reson Med. 2000 Feb;43(2):295-302.

Supervisor(s):

Dr. Andras Jakab, ETF D114.1, Tel.: +41 44 63 27632

Dr. András Jakab, Center for MR-Research, University Children’s Hospital Zürich, Tel.:

Professor:

Orçun Göksel (ogoksel), ETF C107, Tel.: +41 44 63 22529