A New Framework for Mass Detection

Figure  shows a graphical scheme of the proposed framework. The algorithm has been designed as a supervised solution, where the system begins learning a set of parameters from a database of already studied cases. These parameters are the breast tissue, the shape, and the size of the lesions. Moreover, it learns to distinguish between RoIs containing masses and RoIs of normal tissue.

When the algorithm is able to distinguish and classify the breasts/RoIs, it is ready for searching masses in a new mammogram. Roughly, the first step is to know the density of the new breast. Using this information, and the learned shape and size of the masses, it makes use of a Bayesian template matching scheme to find suspicious regions in a mammogram. As those suspicious regions can be a mass or normal parenchyma, a false positive reduction algorithm is applied. This last step is performed using the RoI discrimination learned earlier.

Note that the steps of this framework are the sub-objectives of the thesis. Thus, we have studied and developed a new algorithm to detect masses in a mammogram, another one to make the false positive reduction, and a new algorithm to classify the breasts according to their internal tissue.

We also want to mention here that the proposed template matching algorithm, as well as the false positive reduction scheme, could also be extended and applied to classify other types of mammographic abnormalities, like RoIs containing micro-calcifications or architectural distortions, or even to other types of medical diseases.