diagnosis
Once the lesions are detected, they could be benign or malignant. A further step analysing their morphology is done in conjunction with machine learning techniques to diagnose each case.
We proposed a method for modeling and diagnose microcalcification clusters in mammograms based on their topological properties. The topology of microcalcification clusters is analyzed at multiple scales using a graph-based representation of their topological structure. This method is distinct from existing approaches that mainly concentrate on the morphology of individual microcalcifications and only compute the distance-based cluster features at a fixed scale. In this method, a set of topological features are extracted from microcalcification graphs at multiple scales, and a multiscale topological feature vector is subsequently generated to discriminate between malignant and benign cases.
On the other hand, we also studied an approach for automatic mass diagnosis. The strategy consisted in three main steps. Firstly, region of interests containing mass and background are segmented using a level set algorithm based on region information. Secondly, the characterisation of each segmented mass is obtained using the Zernike moments for modeling its shape. The final step is the classification, which is done using the Gentleboost algorithm that also assigns a likelihood value to the final result.