Computer Assisted Characterization of Lymph Nodes Using Spectral Ultrasound Backscatter and Attenuation Measures

This paper deals with a system for computer assisted characterization of biological tissue using diagnostic ultrasound and its application in the differential diagnosis of lymph nodes. Using standard diagnostic ultrasound equipment, ultrasound radio frequency (RF) data originating from lymph nodes of 24 patients were acquired. 12 patients were proven to have malignant alterations of lymph nodes. The proposed system aims at an automated differentiation between malignant and benign cases. In a first step, spectral ultrasound backscatter and attenuation measures were extracted from diffraction corrected RF data, yielding spatially resolved parameter images. A reduced representation of the measures was found using first order statistics and used as tissue describing features. The features were processed by the classification system. An optimal set of features was chosen by a sequential forward selection algorithm and included 3 features. Classification was performed by total cross validation using a probabilistic neural network. Inputs to the network could be biased, depending on the target class. Thereby, the classifier could be forced to reach an arbitrary sensitivity in detecting positive cases. Thus, receiver operator characteristic (ROC) curves could be determined. The area under the ROC curve was 0.94, proving the potential of the proposed method for differentiating malignant and benign lymph nodes.