Evaluation of a Novel Segment Algorithm on Radiologist's Diagnosis of Breast Masses

摘要

Purpose: To investigate the effect of a novel segment algorithm on radiologists' sensitivity and specificity for discriminating malignant masses from benign masses using breast ultrasound (BUS) imaging. Methods: 510 conventional BUS images acquired from 109 masses were processed by a novel segmentation algorithm. Five radiologists were invited to analyze the original and computerized images independently according to the Breast Imaging Reporting and Data System for Ultrasound. For each lesion, the shape, orientation, margin, boundary, posterior acoustic features and calcification were described and a final assessment of the category was provided. Performances of radiologists with or without computer aid were evaluated by ROC curve analysis. Results: The masses became more obvious after being processed by the segment algorithm. Without using the algorithm, the areas under the ROC curve (Az) of the five radiologists ranged 0.70~0.84. With using the algorithm, the Az increased significantly range 0.79~ 0.88 (P<0.001). The average sensitivities of radiologists' diagnosis increased from 70% to 80%, and the average specificity increased from 72% to 79%. Conclusions: The segmentation algorithm can improve the radiologists' diagnosis performance by reducing the image speckles and extract the mass margin characteristics, and the margin of the masses was the most discordant feature before and after the segmentation algorithm.