Abstract:

BACKGROUND: Ultrasound diagnosis is non-invasive, reproducible and suitable for identification of soft tissue characteristics, which are widely used in the auxiliary detection of breast cancer. However, due to the lack of quantitative criteria, the diagnosis has shown a higher false positive rate with greater subjectivity and uncertainty. The problem is unsolved for clinicians regarding how to improve the diagnostic accuracy and reduce unnecessary biopsies.
OBJECTIVE: To detect breast tumors using ultrasound image feature parameters.
METHODS: Using level set method, 100 breast tumors were segmented and computed with 10 feature parameters including elongation factor, compactness factor, circularity factor and 7 Hu moment parameters.
RESULTS AND CONCLUSION: The scatter graph showed elongation, compactness and circularity points were not superposed but almost concentrated around the same area without good differential diagnosis, and then 7 Hu values were used to analyze local areas around tumors with support vector machine classification. Hu1 and Hu 2 exhibited better outcomes, the accuracy was 88% and 88%, sensitivity was 86% and 90%, specificity was 90% and 86%, positive predictive value was 89.58% and 86.54%, and the negative predictive value was 86.54% and 89.58%, respectively. It suggested that ultrasound image feature parameters analysis provides guidance to differential diagnosis of breast tumors.