Coarse To Refined Kidney Segmentation Leveraging The SDRLSE Algorithm

Accurate and reliable kidney segmentation in CT images provides a solid foundation for the clinical diagnosis and treatment of kidney diseases. Most current kidney segmentation methods are semi-automatic interactive, like Segment Anything Model (SAM), relying on expert input and frequent adjustments. Additionally, there are also some end-to-end automatic segmentation models, but they often lack interpretability and medical knowledge input. Moreover, the low contrast between the kidneys and surrounding organs in CT images, coupled with interference from nearby blood vessels, makes accurate differentiation challenging. This study’s approach focuses on a full pipeline methodology for coarse kidney localization and refined segmentation. Furthermore, we employ YOLOv8 and an improved region-growing algorithm for coarse kidney positioning. We then introduce the Structural Distance Regularized Level Set Evolution (SDRLSE) algorithm. This combines target region contrast and structural similarity to improve the Distance Regularized Level Set Evolution (DRLSE) algorithm, achieving stepwise kidney segmentation based on the contextual connections between CT image sequences, and applied to clinical contrast-enhanced CT images. At last, our pipeline outperforms other segmentation methods, achieving an average DICE coefficient of 0.956 and an IoU coefficient of 0.943.