An Adaptive Image Segmentation Approach for Tumor Region Identification in Ultrasound Images

Identifying and recognizing the tumor region, i.e., region of interest (ROI), in medical imaging always plays a critical role in image-guided drug delivery (IGDD). Recent cutting-edge studies have demonstrated the great potential to leverage ultrasound images in IGDD. However, the effect of interference also results in significant challenges to automatically identify the ROI in ultrasound images, as the state-of-the-art methods usually do not have enough capability of handling such a high level of interference. Thus, the objective of this work is to develop an ultrasound-oriented image segmentation method for accurate and robust ROI identification. To achieve this goal, this study proposed a novel adaptive approach, termed B-CLEAR, through an efficient collaboration framework among gradient-based Boundary detection, feature-based Center Locating, and an Edge-Assisted Region growing algorithm. The capability of this new method is validated by a real-world ultrasound image dataset, which is collected from the experiments of colon tumor treatment. The comparison with conventional segmentation algorithms has also demonstrated the superior performance of the proposed approach for ROI identification in ultrasound images.