General
The Institute of Signal Processing and System Theory (ISS) offers a demanding PÜL, dealing with two cutting-edge pattern-recognition problems. One from the area of medical signal processing and the second from the area of speaker recognition. The PÜL is designed to deepen the knowledge gained in the lecture "Detection and Pattern Recognition". Therefore, it is required that ervery participant attended this particular or a similar lecture to have sufficient theoretical background.
Task I
As shown below, the prostate is a gland in the male reproductive system which oftenly develops cancer. In fact, prostate cancer is the most frequently diagnosed cancer type for males and is one of the leading causes of cancer death among men in the USA.
Non-invasive medical imaging technique like Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) can be used to localize the cancer tissue. Five examples of images obtainted with those two methods are shown below. The prostate of the patient is surrounded by the blue circle, whereas the cancer tissue is surrounded by the red circle. Up to now the classification is conducted manually by experts, what is a time consuming and expensive process.
Participants of this PÜL will work in teams of two, developing a pattern-recognition software for automatic pixelwise classification of prostate tissue, separating healthy prostate tissue from cancer tissue.
Task II
Automated speakers recognition in audio signals can be useful in many situation. Speaker recognition is used to add meta information to audio meeting records for automated storage, to improve service quality and customer satisfaction in telephone hotlines or to improve security in telephone banking applications. In this PÜL the speaker recognition problem shall be examined using the example of recognizing speakers in short audio samples.
Educational objective
This PÜL is an excellent opportunity to get an insight into the fascinating area of medical imaging and to extend your academic knowledge about methods of pattern recognition. The main educational objectives of this PÜL are issues which are not tought in the DPR lecture but which are still important for practical applications:
- Application of training methods to find the best parameters
- Choosing a classifier which is well suited for the given problem
- Feature selection to find subsets of meaningful features
- Methods of feature normalization
- The implementation of simple classifiers
- Dealing with faulty training data (outlier detection/removal)
Scope
The PÜL counts 6 ECTS credits (180 hours of work) and takes place every winter term. Enrollment is in the first lecture week. The information event and first meeting is in the second week. From then on a weekly mandatory meeting is offered until the last week of the leture period, when a short written report has to be submitted and each group has to present their results in a short presentation. Beside the weekly mandatory meeting the software laboratory can be used during regular working hours.
Dates, Registration
please use ILIAS for more information on the dates and the registration (you have to be logged in to ILIAS).
Contact
M.Sc. Lukas Mauch