A new approach to ultrasonic detection of malignant breast tumors

In this work, we report the use of ultrasound RF time series analysis for separating benign and malignant breast lesions with similar B-mode appearance. The RF time series method is versatile and requires only a few seconds of imaging. We have employed the spectral and fractal features of ultrasound RF time series and used support vector machines with leave-one-patient-out cross validation of the classification. We have also produced cancer probability maps, by estimating the posterior malignancy probability of regions of size 1 mm2 in the suspicious lesions. The first 12 patient cases of our ongoing study are reported here. Pathologic analysis of the cores using ultrasound guided needle biopsy confirmed the tissue type. We report an area under receiver operating characteristic curve of 0.82. We were able to successfully classify 6 out of 7 patients with malignant breast lesions and 4 out of 5 patients with benign lesions, with success defined as correct classification of at least 75% of the 1 mm2 regions in the area of the lesion. The above findings suggest that ultrasound time series along with support vector machines can help in differentiating malignant from benign breast lesions.