Neural network based microcalcification detection in a mammographic CAD system

We present the computer aided diagnosis system being developed to help experts in screening mammography. It is a very important project because about 8% of women develop breast cancer in her lifetime therefore global screening is necessary. It means that reliable diagnosis of huge number of images must be solved. The basic architecture of the system and the information processing need is presented. One of the most important tasks in mammographic diagnosis systems is microcalcification detection. It is solved by a hierarchical neural architecture. The original suggestion of that structure was improved by two ways. The image features used as inputs to the neural networks were analyzed and the feature set was extended. The neural architecture was embedded in a neural ensemble context for improving the quality of the solution further. Results of the tests of that detection procedure show that the false detection ratios are acceptable