Three-dimensional (3D) quantitative-ultrasound (QUS) methods were recently developed and successfully applied to detect cancerous regions in freshly-dissected lymph nodes (LNs). The 3D high frequency ultrasound (HFU) images obtained from these LNs contain three different parts: LN-parenchyma (LNP), fat, and phosphate-buffered saline (PBS). To apply QUS estimates inside the LNP region, an automatic and accurate algorithm for LNP segmentation is needed. In this paper, we describe a novel, nested-graph-cut (NGC) method that effectively exploits the nested structure of the LN images. To overcome the large variability of the intensity distribution of LNP pixels due to acoustic attenuation and focusing, we further describe an iterative self-updating framework combining NGC and spline-based robust intensity fitting. Dice similarity coefficients of 89.56±8.44% were achieved when the proposed automatic segmentation algorithm was compared to expert manual segmentation on a dataset consisting of 115 LNs.