VESUNETDeep: A Fully Convolutional Deep Learning Architecture for Automated Vessel Segmentation

Abstract

Segmentation of blood vessels has a lot attention in medical image processing because of its use in the diagnosis of diseases. Although manual segmentation is possible for each patient, it is a laborious and repetitive task which requires professional skills. Thus, many methods have been proposed in the literature, from hand-designed filters to learning-based approaches and machine learning-based approaches outperform hand-designed filters. In this study, we propose an automated, fast and robust deep learning architecture for improving the performance of vessel segmentation. The segmentation performance is compared with the methods in the literature in terms of accuracy, sensitivity, specificity and area under curve (AUC) metrics. Moreover, the proposed deep learning architecture, VESUNETDeep, is tested with and without pre-processing of the input signal. It has been found that the pre-processing step increases sensitivity but decreases a small amount of other metrics. Finally, VESUNETDeep architecture is superior in terms of accuracy, specificity and AUC metrics.