Logo System Theory Group

Signal & System Theory Group

Professor Peter Schreier
Universität Paderborn, Germany
If you are interested in a thesis in statistical signal processing/communications/biomedicine, but can't find a suitable topic below, please feel free to contact us.
Übrigens: Auch wenn ein Thema auf Englisch ausgeschrieben ist, können Sie Ihre Arbeit auf Deutsch schreiben.

Medical image processing of X-ray images

Description

Computer assisted surgery (CAS) has been successfully used in brain and spine surgeries for many years. However, classical CAS systems have some disadvantages (in particular, changed operating room workflow and increased complexities), so that they are rarely used in trauma surgery, where efficiency and simplicity are of the essence.

We are assisting one of the world's leading medical technology companies in their development of a revolutionary CAS system for the treatment of hip fractures, which are breaks in the upper end of the femur (highlighted in the figure on the left). During conventional hip fracture surgery, surgeons use mechanical instruments and X-ray images to place a nail and lag screw inside the femur. Proper positioning of the lag screw in the femoral head is important to for the success of the surgery. An improperly placed lag screw may lead to a "cut out," which necessitates revision surgery.

The new system assists surgeons with implant alignment, appropriate lag screw length selection, and lag screw positioning. This system presents a paradigm shift in computer assisted surgery because it completely adapts to the surgeon and requires little to no interaction between surgeon and system.

This is achieved through automatic evaluation of X-ray images (seen on the right) using intelligent image processing algorithms, which automatically recognize bone structures. In this thesis, you will contribute to the development of these algorithms.

Prerequisites
Knowledge of signal theory is required, a background in image processing is desirable but not essential. Programming skills (MATLAB, Python, or C++) are required. The scope of the project will be adjusted to match the level of offering (bachelor or master).
Contact
Prof. Peter Schreier,

Signal detection with application to Cognitive Radio

Description


Credit: Wang et al., IEEE Journal of Selected Topics in
Signal Processing (2011)
Since the usable radio spectrum is of limited physical extent it is a resource of high demand. To manage its access it is divided into sub-bands and has licenses assigned to them. However, the licensed frequency bands are typically underutilized. In order to improve the wireless spectrum utilization, cognitive radio (CR) is a new communications paradigm that manages the spectrum access dynamically. Hence, unused license bands are accessed by “cognitive” users opportunistically. To provide interference-free communication for all users vacant frequency sub-bands have to be detected reliably.


Credit: Wang et al., IEEE Journal of Selected Topics in
Signal Processing (2011)
A robust spectrum sensing technique is based on cyclostationary (CS) feature detection. Since digital communication schemes produce CS signals, this property can be exploited when deciding whether a given frequency sub-band is occupied. In practice, however, we have to deal with almost-CS signals, which are much trickier to detect. In this thesis, you will look at detectors of (almost-) cyclostationarity. Theoretical questions can include the implementation of multiple hypothesis tests to detect the presence of an (almost-) CS signal or the estimation of the signal's cycle period. Alternatively, practical aspects can be explored by implementing different detectors on a hardware testbed and evaluating their performance under realistic conditions.

Prerequisites
Knowledge of statistical signal processing is essential and a background in digital communications is desirable. Prior experience with MATLAB or Python is helpful.
Contact
Stefanie Horstmann, [javascript protected email address]

Interference Management in Wireless Communications

Description


Credit: Zahir, Talha, et al.,
IEEE communications surveys & tutorials (2013)
Interference from other transmitters is the main bottleneck for most wireless communication systems, which are called interference-limited systems. Due to the bandwidth shortage, these systems must tolerate some interference in order to enhance the spectrum usage. There are different approaches to handle interference in wireless systems. Among them is employing improper signaling. In improper signals, the powers of the real and imaginary parts are not equal, and/or there is correlation between the real and imaginary parts of the signal. The capacity-achieving signal in traditional systems is a proper signal; however, it has been shown that improper signaling can increase the rate in interference-limited systems.


Credit: Yucek, Tevfik, and Huseyin, Arslan,
IEEE communications surveys & tutorials (2009)
In this thesis, you will study the performance of an interference management technique in a wireless communication system. To this end, we first define a scenario in which the system is interference-limited. For instance, we can consider an underlay/overlay cognitive radio (CR) system. In CR systems, licensed or primary users (PU) share the spectrum with unlicensed or secondary users (SU) under the constraint that the PU's communications are not affected by the SU's transmission. In such system, we should design the transmission strategy of SU to get the desired system performance. Moreover, different transmission techniques (e.g., OFDM and single carrier) can be considered in the thesis. You will consider different interference management techniques including improper signaling and interference alignment. Then, you will optimize the performance of the defined scenario by analytical tools. Finally, you will evaluate your results by simulations using MATLAB.

Prerequisites
Knowledge of wireless communications and mathematical tools, i.e., linear algebra and/or optimization methods is required. Depending on your background, scope and focus of the task can be adjusted. Some prior experience with MATLAB is also expected.
Contact
Mohammad Soleymani, [javascript protected email address]

Automatic statistical shape models of the femur bone

Description

Object segmentation within images is an easy task for a human but so much more difficult for a computer/machine. It becomes especially challenging when the object to be segmented belongs to the human body since anatomy has a tremendous variability among individuals. This is the case in Computer Assisted Surgery (CAS) applications for the treatment of hip fracture where the femur bone must be segmented. In this case the task is even harder as the images are generated with a low dose X-ray source. The results are noisy and have low-contrast. Landmarkbased segmentation is a promising approach to solve this problem: given prior knowledge of the position and local appearance of landmarks (right figure), a supervised segmentation method finds the position of these landmarks in new images. In the last couple of decades Active Shape Model algorithm has been widely used for this purpose. It uses a statistical shape model learned from annotated training images (see figure below).

One of the most important steps in the Active Shape Model approach is the annotation of the landmarks in the training images, which is generally done manually. The most common choice is to use anatomical points, for instance “place landmark 24 always in the beginning of the femoral head sphere”. However, this strategy is only based on experience tip and is not optimal in any sense. In this thesis you will use femur shape contours from a set of training images and then develop an algorithm for automatic placement of the landmarks, which is in some sense optimal. The aim is to place points so that the statistical shape model best captures the bone variation among individuals, but with minimal representation error and maximum segmentation performance.

Prerequisites
Knowledge of signal and probability theory; a background in image processing is desirable but not essential. Programming skills (at least MATLAB). A desire to learn and explore new fields is the most important prerequisite. The scope of the project will be adjusted to match the level of offering (bachelor or master).
Contact
Alma Eguizabal, [javascript protected email address]

Fusion of brain imaging data from different modalities

Description


Credit: blogs.scientificamerican.com
In biomedical imaging for the study of brain function, an increasing number of studies are collecting multiple measurements from different modalities, in particular, functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and structural MRI (sMRI). All of these are noninvasive brain imaging techniques: fMRI measures the changes in blood-oxygenation in the brain, EEG measures the brain electrical field through the scalp, and sMRI provides information about the type of brain tissue. These modalities provide complementary information. For instance, fMRI has very good spatial resolution, but bad temporal resolution, whereas EEG has high temporal resolution but poor spatial localization. It is thus of interest to fuse the measurements obtained from these different techniques to combine their respective advantages.

Approaches for data fusion can be classified as either model based or data driven. Model-based approaches require detailed a priori knowledge about the experiment to be performed and the properties of the data. When performing complex experiments, the underlying dynamics become very difficult to model, in which case data-driven analysis methods (e.g., correlation analysis techniques such as CCA) are to be preferred. In this thesis, you will investigate different data-driven techniques for making group inferences. For instance, we might be interested in analyzing fMRI data from the same subjects scanned at different alcohol levels while performing a given task.

Prerequisites
Knowledge of statistical signal processing and linear algebra. Experience with MATLAB is helpful.
Contact
Tanuj Hasija, [javascript protected email address]

Improper signaling for interference management in multiuser wireless networks

Description

Due to the broadcast nature of the wireless medium, interference between users presents the major limitation in wireless communications. Typically, interference is avoided by assigning each user a different time-slot or frequency band (TDMA and FDMA, respectively). From an information theoretic standpoint, however, the performance could be substantially improved by interference coordination techniques. Such techniques are based on user cooperation, whereby transmitters adjust their parameters (transmit power, transmit directions, etc.) exploiting information about the other users (typically the channel responses, which must be previously acquired), with the goal of minimizing the impact of the interference.

Among the different interference coordination techniques, improper signaling, which is the transmission of signals with correlated in-quadrature and in-phase signals, is increasing in popularity due to the fact that it can be used in conjunction with other techniques and without incurring any additional cost. In this thesis, you will analyze the benefits of improper signaling for different relevant multiuser networks. A special focus will be devoted to the analysis and development of algorithms for both single-antenna and multiple-antenna systems.

Prerequisites
It is essential to have some knowledge of linear algebra and communications. Some prior experience with Matlab is also important.
Contact
Christian Lameiro, [javascript protected email address]

Unsupervised attribute discovery

Description

Pattern recognition tasks attempt to classify objects into meaningful categories. For example, the postal service might like to classify images of handwritten digits, such as those in Figure 1, into groups based on the numbers that they represent to make sorting the mail easier. The police might want to classify mobile phone videos based on the scene in which they were filmed in order to localize the search for a suspect. Many methods for pattern recognition rely on the extraction of features, or measurable properties of the object, that allow them to distinguish object categories. That is, the methods find aspects of the data, like visual edges in images or foreground movement in videos, that help identify what is distinct about the properties of interest in the data. Unfortunately, these features are often difficult to interpret, e.g. the Gabor-like texture filters generated by a convolutional neural network as in Figure 2. These filters are very useful for image classification, but knowing that a particular filter has a strong response to an image tells us very little about the content of that image. On the other hand, the characteristics that humans use to describe a class of objects might not be useful for separating a particular set of data, or may take a great deal of effort to leverage. For example, someone might describe dogs as scruffy, loyal, and loving, but those traits don't make it easier for a computer to identify dogs.


Credit: Kaparthy et al. CVPR 2014
In this light, we would like to find features that are simultaneously semantically meaningful and useful for classification with limited or no human supervision. We will attempt to automatically extract interpretable attributes from the data using correlation between objects that are represented by linear combinations of a handful of distinct images. We expect that samples will demonstrate dependency in the spatial domain and potentially in other domains as well. Thesis topics could relate to improving estimates of multiset canonical correlation when the samples are known to be dependent, designing and evaluating a subspace median to generate a visual exemplar of related objects, or hypothesis testing for determining when a set correlated objects are different enough from the collection to represent an attribute.

Prerequisites
Knowledge of linear algebra, and some familiarity with Matlab or Python are expected. Experience with image processing would be helpful but is not required.
Contact
Tim Marrinan, [javascript protected email address]