Assessing algorithms for defining vascular architecture in subharmonic images of breast lesions

The ability to accurately and non-invasively characterize breast lesions and their vasculature would greatly limit the number of unneeded biopsies performed annually. Subharmonic ultrasound imaging (SHI) allows exclusive imaging of vasculature in real time, while completely suppressing tissue signals. Previously, cumulative maximum intensity (CMI) projections of SHI data were shown to be useful for characterization, but lacked means of quantification. In this study we investigate three potential thinning algorithms for defining breast lesion architecture. Sequential thinning, parallel thinning, and distance transformation algorithms were compared using 40 in vitro test images. Sequential thinning was selected due to superior connectivity, minimal rotational variance, and sufficient data reduction. This algorithm was then applied to 16 CMI SHI images of breast lesions, out of which 13 were successfully skeletonized. Average bifurcations were 9.8 ± 8.18 and 6.9 ± 6.50 in malignant and benign lesions, respectively (p > 0.60). Average vessel-chain length was 88.9 ± 79.10 pixels versus 63.2 ± 45.65 pixels in malignant versus benign lesions (p > 0.40). While the sequential thinning algorithm was promising for quantifying breast vasculature, its ability to significantly differentiate between malignant and benign lesions in this study was limited by a high degree of variability and limited sample size.