Analyzing Tree-Like Structures in Biomedical Images Based on Texture and Branching: An Application to Breast Imaging

摘要

We propose an approach for analyzing tree-like structures in biomedical images. Our analysis is based on Vector Quantization (VQ), an image compression technique. Here, we approach VQ from a different perspective: we use the histogram of the codeword usage as a feature vector representing the initial image. As ductal tree topology has predictive value for a variety of diseases, such as papilloma, ductal ectasia, and ductal carcinoma, we chose to apply this technique to compare texture of the breast ductal tree in X-ray galactograms against the same tissue in corresponding unenhanced mammograms, which do not visualize the ductal tree. We also investigate the relationship between texture and the underlying ductal branching topology using descriptors adapted from the data mining literature. We believe that our method has the potential to assist the interpretation of clinical images and deepen our understanding of relationships among structure, texture, function, and pathology.