In this thesis, we have developed and researched a medical multimedia system with the focus on automatic analysis of gastrointestinal (GI) diseases. The system is called Eir, after the Norwegian goddess of healing, and has a focus on accurate and time efficient processing of multimedia data. We investigated parallel processing, GPU-based acceleration and different classification approaches that are evaluated and compared with state-of-the-art methods such as deep learning and artificial intelligence.
Eir combines search-based classification based on information retrieval methods and machine learning to be both fast and accurate at the same time. Then, using real-world datasets and in close collaboration with the Cancer Registry of Norway, Bærum Hospital and the Karolinska Hospital we have demonstrated that the Eir system can outperform state-of- the-art approaches in both automatic analysis and processing speed, reaching detection accuracy above 93% and processing speed above 300 frames/images per second, respectively. With our promising results we could attract several hospitals for collaboration, and the Eir system is momentarily prepared for being tested and used under clinical conditions within our collaborating hospitals in Norway, Italy, Japan, Sweden and USA
Health care systems all over the world have a long history of adopting technology for being able to improve care, quality of life and patient survival. Visual information is frequently used to support medical experts in their daily tasks, such as disease detection and analysis. In this context, the human digestive system can potentially be affected by many types of diseases. For example, three of the six most common cancer types are located in the GI tract, with about 2,8 million new cancers yearly and a mortality of about 65%. Early detection is critical for the outcome, level of treatment and survival of the patients. Common diseases in the GI tract, beside colorectal cancer include gastroesophageal reflux disease, peptic ulcer disease, inflammatory bowel disease, celiac disease and chronic infections. Norway has one of the highest incidences of colorectal cancer worldwide, and the numbers are increasing. All possible GI diseases have a significant impact on the patients’ health-related quality of life. Consequently, gastroenterology is one of the biggest medical branches, and the potentialimpact of the presented research is therefore large where early detection of abnormalities greatly increases the chance of successful treatments.
The thesis is written within the field of Communication Systems. The work has been conducted at Simula Research Laboratory.
Prior to the defense, at 10.15, Michael Riegeler presented his trial lecture "Overview of Variants of Deep Learning for Visual Data".
The adjudication committee
- Associate Professor Roger Zimmermann, School of computing, Department of Computer Science, National University of Singapore, Singapore
- Professor Christian Breiteneder, Institute of Software Technology and Interactive Systems, Vienna University of Technology, Austria
- Professor Jim Tørresen, Department of informatics, University of Oslo
Chair of the disputation
- Ellen Munthe-Kaas,Head of Department, Department for Informatics
- Professor Pål Halvorsen,Department of Informatics, University of Oslo
- Professor Carsten Griwodz,Department of Informatics, University of Oslo