Speckle reduction approach for breast ultrasound image and its application to breast cancer diagnosis

Abstract

Objectives: To retrospectively evaluate the effects of a speckle reduction algorithm on radiologists' diagnosis of malignant and benign breast lesions on ultrasound (US) images.

Methods: Using a database of 603 breast (US) images of 211 cases (109 benign lesions and 102 malignant ones), the original and speckle-reduced images were assessed by five radiologists and final assessment categories were assigned to indicate the probability of malignancy according to BI-RADS-US. The diagnostic sensitivity and specificity were investigated by the areas (Az) under the receiver operating characteristic (ROC) curves.

Results: The sensitivity and specificity of breast lesions on Ultrasound images improved from 88.7% to 94.3%, from 68.6% to 75.2%, respectively, and the area (Az) under ROC curve of diagnosis also increased from 0.843 to 0.939, Z=4.969, there were significant differences in the Az between the original breast lesions and speckle-reduced ones on Ultrasound images (P<0.001). The diagnostic accuracy of breast lesions had been highly improved from 78.67% to 92.73% after employing this algorithm.

Conclusions: The results demonstrate the promising performance of the proposed speckle reduction algorithm in distinguishing malignant from benign breast lesions which will be useful for breast cancer diagnosis.