Vessel enhancement with multiscale and curvilinear filter matching for placenta images

Recently, placental pathology evidence has contributed to current understanding of causes of low birth weight and pre-term birth, each linked to an increased risk of later neuro-developmental disorders. Among various factors that cause such disorders, the vessel network on the placenta has been hypothesized to offer the most clues in bridging that connection. Herein lies the most essential step of the blood vessel extraction, which has only been done manually through laborious methods. In this paper, a filtering process that is partly based on images' second-order characteristics is proposed to highlight image pixels from locally curvilinear structures while simultaneously decrease non-vessel noise. Results are reported in Matthews Correlation Coefficient (MCC) against the pathologist's ground truth tracings and compared with an existing neural network approach. The proposed enhancement process consistently outperforms the multiscale and neural network approaches in both accuracy and efficiency. Since the process is completely automated, the algorithm is readily extendable to other medical images where vessel extraction is needed.