Spatial feature extraction techniques for the analysis of ductal tree structures

The objective of this study is the classification of the breast ductal tree structures appearing in galactograms in order to investigate the relation between topological properties of the tree-like parenchymal networks and radiological findings regarding breast cancer. We present two different methods to characterize the spatial distribution of branching points; a variation of Sholl´s analysis and a sectoring technique. Similarity searches and k-nearest neighbor classification of the trees are performed using the cosine similarity metric. We applied our approach to clinical x-ray galactograms to distinguish between cases with reported radiological findings and cases with no reported findings. Experimental results illustrate the effectiveness of the proposed methods, indicating that our analysis can potentially aid in early breast cancer diagnosis.