Abschlussarbeiten

Im Projekt müssen grosse Mengen an medizinischen Daten in einem DataWarehouse gespeichert (SQL-Datenbanken) werden, mit Indizees indexiert (Lucene, Solr, ElasticSearch, Neo4J) werden und von Benutzern abgefragt werden können (grafische Webfrontends).

Zusätzlich müssen auf den textuellen Dokumenten der Daten Informationsextraktion durchgeführt werden um diese Daten strukturell indexierbar zu machen (Information Extraction).

Wenn dich das Projekt oder seine technischen Themenfelder interessiert melde dich bei georg.fette (at) uni-wuerzburg.de

If you are interested in image processing and if you are looking for a topic for your bachelor/master thesis or your practical course, please contact Amar Hekalo, Adrian Krenzer or Prof. Frank Puppe. You can find an overview of our general research interests as well as some concrete topics below. If not mentioned otherwise, each topic is available as bachelor thesis, master thesis and practical work.

Medical Image Processing (in general)

Medical image processing is an important and rapidly growing field of research. It can be separated into two main sub-areas, we would like to supervise studential works in:

• Segmentation is the process of detecting regions of interest and usually also separating them from the rest of the image. In the example below (left) the heart (red), lung (green) and clavicle (blue) have been identified in a radiograph of the thorax.
• Classification often follows the segmentation step. Its goal is to categorize an object using its properties, so-called features. For example, a lesion identified in a mammography (right) can be classified as benign or as malignent (red).

If you are interested in any kind of topics regarding medical image segmentation or classification, please don´t hesitate to contact us, even if there are currently no theses topics listed below. Furthermore, you are always welcome to approach us with your own ideas for the topic of your thesis. All you need is a medical image dataset, including the corresponding gold-standard/ground truth data. Many datasets are publicly available, for example at grand-challenge or the Medical Image Database of the Cornell University.

Image Detection in Scans

To optimize the results achieved by OCR (Optical Character Recognition) methods on a scanned page, a preceding analysis of the layout is needed. This also includes identifying and segmenting images (if present).

Freely available tools like Tesseract often struggle with this task. Therefore, the first step during this work will be to research several image detection algorithms. Next, some of the algorithms have to be tested and evaluated on different domains. Depending on the results, a new algorithm has to be developed by making adjustments, like for example using prior knowledge about the present domain.

Image Matching

"Schulwandbilder" are large-sized images which were used in class-rooms to vividly impart learning content. Accompanying booklets like "Der praktische Schulmann" and "Der neue Schulmann" provide additional information about the images as well as recommendations of how to optimally use them in class.

The goal of this work is to find an efficient way to accurately match the original images (left) to the ones pictured in the accompanying booklets (right). This isn´t a trivial task as even matching images can differ greatly regarding size, color and scan quality.