Institute of Modern Optics, College of Information Technical Science, Nankai University, Key Laboratory of Optical Information Science and Technology, Ministry of Education, Tianjin 300071, China.
Med Phys. 2013 Aug;40(8):081905. doi: 10.1118/1.4812428.
Accurate segmentation of renal tissues in abdominal computed tomography (CT) image sequences is an indispensable step for computer-aided diagnosis and pathology detection in clinical applications. In this study, the goal is to develop a radiology tool to extract renal tissues in CT sequences for the management of renal diagnosis and treatments.
In this paper, the authors propose a new graph-cuts-based active contours model with an adaptive width of narrow band for kidney extraction in CT image sequences. Based on graph cuts and contextual continuity, the segmentation is carried out slice-by-slice. In the first stage, the middle two adjacent slices in a CT sequence are segmented interactively based on the graph cuts approach. Subsequently, the deformable contour evolves toward the renal boundaries by the proposed model for the kidney extraction of the remaining slices. In this model, the energy function combining boundary with regional information is optimized in the constructed graph and the adaptive search range is determined by contextual continuity and the object size. In addition, in order to reduce the complexity of the min-cut computation, the nodes in the graph only have n-links for fewer edges.
The total 30 CT images sequences with normal and pathological renal tissues are used to evaluate the accuracy and effectiveness of our method. The experimental results reveal that the average dice similarity coefficient of these image sequences is from 92.37% to 95.71% and the corresponding standard deviation for each dataset is from 2.18% to 3.87%. In addition, the average automatic segmentation time for one kidney in each slice is about 0.36 s.
Integrating the graph-cuts-based active contours model with contextual continuity, the algorithm takes advantages of energy minimization and the characteristics of image sequences. The proposed method achieves effective results for kidney segmentation in CT sequences.
准确分割腹部 CT 图像序列中的肾组织是计算机辅助诊断和临床应用中病理检测的不可或缺步骤。本研究旨在开发一种放射学工具，以提取 CT 序列中的肾组织，用于管理肾诊断和治疗。
在本文中，作者提出了一种新的基于图割的主动轮廓模型，该模型具有用于 CT 图像序列中肾提取的自适应窄带宽度。基于图割和上下文连续性，逐片进行分割。在第一阶段，根据图割方法交互式地分割 CT 序列中的中间两个相邻切片。随后，通过所提出的模型，使可变形轮廓向肾边界演化，以提取其余切片中的肾。在该模型中，边界与区域信息相结合的能量函数在构建的图中进行优化，自适应搜索范围由上下文连续性和对象大小确定。此外，为了降低最小割计算的复杂性，图中的节点仅具有 n 个链接，从而减少了边的数量。
使用 30 组正常和病理性肾组织的 CT 图像序列来评估我们方法的准确性和有效性。实验结果表明，这些图像序列的平均骰子相似系数从 92.37%到 95.71%不等，每个数据集的相应标准差从 2.18%到 3.87%不等。此外，每个切片中一个肾脏的自动分割平均时间约为 0.36 秒。
该算法将基于图割的主动轮廓模型与上下文连续性相结合，利用能量最小化和图像序列的特点。所提出的方法在 CT 序列中的肾分割中取得了有效的结果。
