A Tumor Detection Algorithm for Whole Breast Ultrasound Images Incorporating Breast Anatomy Information

A whole breast ultrasound tumor detection algorithm, which could help physicians determine and diagnose, has been proposed. First of all, we employ the multi-scale blob detection. Secondly, we discard most unlikely lesions by ultrasound confidence maps and sheet detection, according to the prior knowledge of breast anatomy. After discarding unlikely blob structures, we collect the survival blob-like structures from the results of four passes of multi-scale blob detection in a single set. The features of blobness, size, and probability of being at muscle layer of the all detected blob-like candidates are used in a classification process for the differentiation of true lesions from negative ones in the collection set. Mutual information based feature selection (MIFS) procedure is applied to select three most effective features for the purpose of dimension reduction with the aid of logistic regression classifier and the process of leave-one-out cross-validation (LOO-CV) method. 49 datasets were acquired from 29 patients, which contain 86 lesions in total. According to the area under the ROC curve (AUC), the technique shows promising future for computer aided detection and the superior results when compared to Moon's method.