Xiaohong Wang, Abstract Co-Author: Nothing to Disclose

Weijun Peng MD, Abstract Co-Author: Nothing to Disclose

Haibin Huang MD, PhD, Abstract Co-Author: Nothing to Disclose

Xiao Jie, Abstract Co-Author: Nothing to Disclose

Reto Dominik Merges, Presenter: Employee, Siemens AG, Shanghai, China

To investigate the value and feasibility of an automated volume segmentation software in the assessment of lymphoma lesions in different locations and of different sizes.

60 lymphoma lesions were analysed in this study(less than 5 per patient). All were examined using routine Muli-Slice Computed Tomography (SOMATOM Sensation 40; Siemens AG Medical Solutions, Germany).The axial images of interested imagings were reconstructed with 2mm slice thickness. Then 2D and 3D reconstruction were done on a post processing workstation. The segmentation algorithms implemented in Prototype version of syngo CT Oncology software(Siemens AG Medical Solutions,Germany) were measured against ground truth data, established via manual contouring by experienced radiologists. For the evaluation of the effect of lesion location and size on the accuracy, metrics were assessed using Kruskal-Wallis test and multiple comparisons.

(1) Lymphoma lesions we researched were located in chest, neck, and abdomen. Three metrics were used to evaluate the accuracy performance of the automated segmentation software with the results in the respective locations(chest, neck, abdoman): volume overlap (64%, 72%, 68%), over-segmentation ratio (8.5%, 5.8%, 5.3%) and under-segmentation ratio (12.3%, 10.1%, 10.8%). (2) Results from Kruskal-Wallis test and multiple comparison demonstrated that lesions in the neck region showed largest volume overlap (P0.05). (3) the algorithm was robust (insensitive) to lesion size, demonstrated by the small coefficient of determination (close to 0) in the linear regression analysis.

The tested segmentation algorithms are accurate for the evaluation of the volume of lymphoma lesions regardless of the location, especially for the lesions in the neck. The algorithms are insensive to the size of the lymphoma.

The used automatic volume segmentation software is feasible and a useful and effective tool to evaluate lymphoma lesions and should be used in a wide range of cases.

Wang, X, Peng, W, Huang, H, Jie, X, Merges, R, Lymphoma Lesions: Effect of Location and Lesion Size on the Accuracy of an Automated Volume Segmentation Algorithm.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5010132.html