Breast Cancer Detection from a New Ultrasound RF Time Series Based Approach: Phantom Studies

In breast cancer which is a women common cancer, the treatment will be more successful in early stages. Due to the lack of sufficient accuracy of each nowadays method to determine the type of lesions, a biopsy is used. But it's invasive and causes a lot of mental and financial costs on the patient. Thus, it is necessary to contemplate a non-invasive and non-radioactive method to ensure about the type of lesion. Researches on echo patterns of radiofrequency RF ultrasound data have shown that it had information about the propagation environment and can be used for tissue typing. In another work we suggested extracting RF time series features in the progress of stimulating breast tissue added more information to the RF time series and improve tissue classification results. Data were recorded from two types of agar-gelatin phantom mimicking breast tissue. Data collection was performed under a fixed probe, static and vibrational excitation. Finally, SVM classifiers have been used to classify healthy and cancerous tissues. In this study, the effect of some recording parameters (depth and probe central frequency) on this classification has also been investigated. A Supersonic Imagine US system equipped with a linear probe was used to record this signal. The most improvement in classification accuracy was achieved by using vibration stimulation and 6.4 central frequency at 2cm depth. This result is for ultrafast beamformed data with
$98.78\pm 2.77$
% accuracy.