This Article Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Zhang, J. Articles by Wang, Y. Search for Related Content PubMed PubMed Citation Articles by Zhang, J. Articles by Wang, Y.

Sonographic Feature Extraction of Cervical Lymph Nodes and Its Relationship With Segmentation Methods

Electronic Engineering Department, Fudan University, Shanghai, China (Jh.Z., Yy.W.); and Ultrasound Department, Huashan Hospital, Shanghai, China (Y.D., Y.W.).

Address correspondence to Yuanyuan Wang, PhD, Department of Electronic Engineering, Fudan University, Shanghai 200433, China. E-mail: yywang{at}fudan.edu.cn

Objective. The purpose of this study was to extract quantitative features for characterization of cervical lymph nodes on sonographic images and analyze the effect of a semiautomated segmentation method on the feature extraction. Methods. Contours of 186 cervical lymph nodes on sonographic images were separately delineated by 2 radiologists (R1 and R2) and a semiautomated segmentation method. For each node, 10 kinds of sonographic features (including 3 parameters of size; 12 parameters of margin; 4 parameters of nodal border; 10 parameters of echogeneity; and 1 parameter of shape, echogenicity, medulla ratio, medulla distribution, vascular density, and vascular pattern, respectively) were quantified by a computerized scheme based on the segmented contour. Correlations between the quantitative parameter and the radiologists’ consensus grading were computed to assess the effectiveness of these parameters. Concerning the 14 best correlated parameters, the effect of the segmentation stage on the feature extraction was estimated by comparing the parameter values calculated under different segmentations in terms of relative ultimate measurement accuracy. Results. Good correlations between the computerized scheme and radiologists were seen in features of size, nodal border, shape, echogenicity, medulla ratio, medulla distribution, vascular density, and vascular pattern, whereas 10 of 12 parameters of margin features and 8 of 10 parameters of echogeneity features showed poor correlations. Paired t tests comparing the relative ultimate measurement accuracy computed using the R1-R2 and the R1-computer pairing showed no significant difference on 11 parameters for the 14 parameters analyzed. Conclusions. The computerized feature parameters may be used as assisted indices for evaluating cervical lymphadenopathies from sonographic images. The semiautomated segmentation method satisfied the accuracy requirement of the feature extraction.

Key Words: cervical lymph node • feature extraction • segmentation • sonographic image

Abbreviations: AMAG, average maximum ascending gradient • AMDG, average maximum descending gradient • AMINDIST, average minimum euclidean distance • CAD, computer-aided diagnosis • EDV, edge distance variation • EPPD, entropy of the projection probability distribution • FD, Fourier descriptor • GLN, gray level nonuniformity • GVF, gradient vector flow • L, long axis • NRL, normalized radial length • RLN, run length nonuniformity • RUMA, relative ultimate measurement accuracy • S, short axis • VI, edge intensity variation