Support vector machines based computer-aided diagnosis system of breast tumor with ultrasound images

A computer-aided diagnosis system of breast tumor with ultrasound images based on is proposed. Based on the criterions used by physicians, total twenty-six features about textures and geometric feature including autocorrelation coefficients, mean square deviation coefficients, and compact ratios are extracted. These features reflect the essential differences between malignant breast tumors and benign tumors, and concern about the characters of ultrasound images as the coherent images. Experiment results show that the classification accuracy is 93.03%, sensitivity is 94.30%, specificity is 91.59%, the positive predictive value is 92.80%, negative predictive value is 93.33%, the areas under the ROC curve is 0.9669.