The goal of much research in medical imaging is to improve the diagnosis made by a radiologist. This means that new techniques and new imaging systems that are developed must prove that they can improve radiologists' ability to render an accurate diagnosis. Observer studies are useful because provide evidence as to the clinical effectiveness of a technique. However, observer studies are difficult to perform, because they are time consuming, and they are expensive to conduct. We are proposing a novel approach of using a computer-aided diagnosis (CAD) scheme to replace the human observer. That is, we believe that we can modify CAD schemes so that, at least for some tasks, they can perform like human observers. In this way, observer studies can be performed using a computer instead of a group of radiologists, saving time, money, and effort. The goal of this project is to use different training conditions to create a set of CAD schemes for the characterization of breast calcifications on mammograms. Then to select a subset of these to create a set of CAD-observers that has similar performance to a group of radiologists.
The specific aims are: 1. Assemble database and perform observer study; 2. Develop superset of CAD-observers; 3. Select a set of CAD-observers to match radiologists; and 4. Validate CAD-observers. If we are successful, many new developments in medical imaging can be tested more accurately, easily, and quickly saving many months and many dollars in development time and costs. This should reduce the cost and the time it takes new technologies or techniques to gain clinical acceptance.
