Automatic slice selection and diagnosis of breast strain elastography.

PURPOSE Generally speaking, breast imaging experts and physicians select a representative slice from the strain elastographic image sequences to diagnose the tumor. Given the strain image qualities, it is difficult to make a successful diagnosis using human eyes only. The main purpose of this study is to develop an automatic and reliable method to select the representative slice from the elastography cine loops and/or video and then diagnose the tumor by means of the elastographic features generated from the selected slice. METHODS In this study, the authors collected 80 biopsy-proven breast tumors, comprising of 45 benign and 35 malignant lesions, to estimate the performance of the automatic slice selection method. Images chosen using several slice selection criteria (e.g., whole-image analysis or tumor region analysis) were compared to the physician-selected images to determine the best selection criterion. The level set tumor segmentation method was applied to the corresponding B-mode part of the representative elastographic slice to overlap tumor boundaries on strain images and to calculate elastographic features for diagnosis. RESULTS The experiment showed that the diagnostic performance, in terms of accuracy, sensitivity, and specificity, evaluated by the leave-one-out method, based on the elastographic features for the representative slice selected by the proposed slice selection method, was 71.3%, 91.4%, and 55.6%, respectively, while the performance values for the physician-selected slice were 65.0%, 77.1%, and 55.6%, respectively. CONCLUSIONS Both the sensitivity and accuracy of the proposed slice selection method were better than those of the physician-selected slice, and the specificity of these two different schemes is similar. According to the statistical analysis of experimental results, the performance of the proposed slice selection method was similar to that of the physician's selection. The authors concluded that the proposed slice selection method could assist the physician in selecting the appropriate representative slice and in decreasing the time of selection.