MICA: 3D Medical Image Collaborative Annotator
Edgar Kwen-Wei LiuOral Defence Date:
Friday, June 13, 2008 - 11:30Location:
Professors Okada, Petkovic, and Yoon
ABSTRACT Medical image analysis is a field that studies how best to process 3D images created by computer-aided medical imaging devices such as CT, MRI and PET. Typically, scanned images are checked by Board-certified radiologists (BCRs), the only valid source of authoritative knowledge, in order to determine whether or not there are lesions on the scanned images. However, these medical imaging devices can generate thousands of scanning results in a short time. Detecting lesions in medical images with machine learning mechanism can greatly reduce the workload of Certified Radiologists. However machine learning mechanism still needs a large amount of good samples to generate better results. It is time-consuming to create samples that are representative of a target problem. Currently we have to rely on the few Radiologists’ knowledge to prepare good samples but they can’t waste their precious time. On the other hand, there are many Non-Experts who are often not limited in their time resource but with a little skill on medical image analysis. What if we can train these Non-Experts and have them generate samples for automated tumor detection? In this paper we introduce a web-based system which helps Non-Experts to get better trained through an asynchronous annotator/validator schema. This system allows Non-Experts and Radiologists to work collaboratively regardless of where they are located or when they are available to work. Our work is also aiming at providing a friendly UI which is easy to understand and operate to users.