Comparison of Conventional US and Spatial Compound Imaging in Diagnostic Performances of Computer Aided Diagnosis System

To compare the diagnostic performances between two computer aided diagnosis (CAD) systems, this research trains one system by conventional ultrasound (US) and the other by spatial compound imaging, for differentiating between benign and malignant breast masses. The study obtains conventional US and spatial compound images in 128 patients with 140 masses including ninety-seven benign and forty-three malignant masses simultaneously. Each mass was characterized by eight computerized features including the morphological class (shape, orientation, and margin) and acoustic class (lesion boundary, echo pattern, and posterior shadowing). The study trained the binary logistic regression model by correlating the pathological results with the computerized features as a CAD system. When utilizing acoustic class features to construct CAD systems, the area indices under ROC curves were 0.80 and 0.85, respectively, using conventional ultrasound and using spatial compound imaging. The area indices were 0.95 and 0.97, respectively, using all features in conventional US and that in spatial compound imaging. The diagnostic performance of the CAD system did not significantly improved using spatial compound imaging.