Michael Schwier, Presenter: Nothing to Disclose

Teodora Chitiboi, Abstract Co-Author: Nothing to Disclose

Andre Homeyer, Abstract Co-Author: Nothing to Disclose

Horst Karl Hahn PhD, Abstract Co-Author: Nothing to Disclose

1. Understand the difficulty of getting computers to automatically identify structures in (medical) images, while it seems rather easy for humans. 2. Understand the concept of object-based image analysis (OBIA) as a new image analysis approach. Most importantly understand the limitations of pixel-based approaches, which are most common in medical image analysis, and how OBIA extends the possibilities for image analysis. 3. Emphasize the importance of medical experts’ contextual knowledge and how OBIA allows the utilization of this for developing better detection methods. 4. Insights into example solutions. Some published and evaluated. Some early experimental to demonstrate how in cooperation with clinicians first prototypes can be developed quickly.  

1. Advantage of human vision – why is it so hard to teach computers to “see”? 2. Idea of object-based image analysis 3. Advantages of object-based over pixel-based image analysis 3a. Features matter – how objects exhibit a wealth of information and pixels don’t 3b. Context matters – how context matters to identify a target object 3c. How to include (contextual) knowledge provided by medical experts 4. Example solutions 4a. Detection of liver lesions (CT) 4b. Vertebrae detection (CT) 4c. Spinal cord detection (MR) 4d. (possibly more depending on space)  

http://abstract.rsna.org/uploads/2014/14018081/14018081_lwd6.pdf

Schwier, M, Chitiboi, T, Homeyer, A, Hahn, H, Utilizing Contextual Knowledge for Computer Aided Detection with Object-based Image Analysis.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14018081.html