Computer-aided diagnosis of peripheral soft tissue tumors using geometric and texture features

In recent years, researchers have investigated different methods for the differentiation of benign and malignant tumors. It, however, is still very limited for the peripheral soft tissue tumors, due to the rare occurrence and large variation of the characteristics. Computer-aided diagnosis (CAD) system makes tremendous progress in the studies of ultrasound images, so it can provide a stable and reliable diagnostic modality in clinic. In our experiment, we studied 49 patients with peripheral soft tissue tumors and verified them with histopathology examination. All patients were examined by the same ultrasound system and linear transducer. Before quantifying analysis, tumor contours are segmented by an active contour model. Following, we retrieve geometric and texture features from ultrasound images to develop a CAD algorithm. Linear discriminant analysis (LAD) with leave-one-out cross-validation is used in the analysis. Both the geometric and texture features are performed the results with p-value < 0.01. After the combination of both two features, we obtain a better result, with accuracy: 93.9%, specificity: 93.8%, sensitivity: 94.1%, and the combination of 7 features are area, echo, energy, homogeneity, correlation, sum entropy, sum variance.