Three-dimensional segmentation of high-frequency ultrasound echo signals from dissected lymph nodes

Quantitative high-frequency ultrasound (QHFU) imaging methods are under investigation to evaluate their ability to detect small nodal metastases in lymph nodes freshly dissected from cancer patients. To apply these methods, a critical preprocessing step is 3D segmentation of the lymph-node ultrasound echo-signal dataset. Segmenting the residual fat layer and the lymph node is critical in order to avoid bias in the QHFU estimates (e.g., scatterer size and acoustic concentration) due to attenuation and to exclude estimates obtained from the fat regions. Segmentation also provides absolute measurements of lymph-node dimensions that are necessary to match 3D ultrasound with 3D histology. In this study, a 3D region-based segmentation algorithm was developed and compared quantitatively using Dice's mutual-overlap criterion with 2D manual segmentation of 9 representative cross sections. The method was tested on 13 lymph nodes, and resulting Dice scores had mean values of 0.81 and 0.78 for lymph node and fat segmentation, respectively.