We propose a framework for localization and classification of masses in breast ultrasound (BUS) images. We have experimentally found that training convolutional neural network based mass detectors with large, weakly annotated datasets presents a non-trivial problem, while overfitting may occur with those trained with small, strongly annotated datasets. To overcome these problems, we use a weakly annotated dataset together with a smaller strongly annotated dataset in a hybrid manner. We propose a systematic weakly and semi-supervised training scenario with appropriate training loss selection. Experimental results show that the proposed method can successfully localize and classify masses with less annotation effort. The results trained with only 10 strongly annotated images along with weakly annotated images were comparable to results trained from 800 strongly annotated images, with the 95% confidence interval of difference -3.00%–5.00%, in terms of the correct localization (CorLoc) measure, which is the ratio of images with intersection over union with ground truth higher than 0.5. With the same number of strongly annotated images, additional weakly annotated images can be incorporated to give a 4.5% point increase in CorLoc, from 80.00% to 84.50% (with 95% confidence intervals 76.00%–83.75% and 81.00%–88.00%). The effects of different algorithmic details and varied amount of data are presented through ablative analysis.

The aim of this study was to compare the performance of image analysis for predicting breast cancer using two distinct regression models and to evaluate the usefulness of incorporating clinical and demographic data (CDD) into the image analysis in order to improve the diagnosis of breast cancer. This study included 139 solid masses from 139 patients who underwent a ultrasonography-guided core biopsy and had available CDD between June 2009 and April 2010. Three breast radiologists retrospectively reviewed 139 breast masses and described each lesion using the Breast Imaging Reporting and Data System (BI-RADS) lexicon. We applied and compared two regression methods-stepwise logistic (SL) regression and logistic least absolute shrinkage and selection operator (LASSO) regression-in which the BI-RADS descriptors and CDD were used as covariates. We investigated the performances of these regression methods and the agreement of radiologists in terms of test misclassification error and the area under the curve (AUC) of the tests. Logistic LASSO regression was superior (P<0.05) to SL regression, regardless of whether CDD was included in the covariates, in terms of test misclassification errors (0.234 vs. 0.253, without CDD; 0.196 vs. 0.258, with CDD) and AUC (0.785 vs. 0.759, without CDD; 0.873 vs. 0.735, with CDD). However, it was inferior (P<0.05) to the agreement of three radiologists in terms of test misclassification errors (0.234 vs. 0.168, without CDD; 0.196 vs. 0.088, with CDD) and the AUC without CDD (0.785 vs. 0.844, P<0.001), but was comparable to the AUC with CDD (0.873 vs. 0.880, P=0.141). Logistic LASSO regression based on BI-RADS descriptors and CDD showed better performance than SL in predicting the presence of breast cancer. The use of CDD as a supplement to the BI-RADS descriptors significantly improved the prediction of breast cancer using logistic LASSO regression.

To evaluate the value of the computer-aided diagnosis (CAD) program applied to diagnostic breast ultrasonography (US) based on operator experience. US images of 100 breast masses from 91 women over 2 months (from May to June 2015) were collected and retrospectively analyzed. Three less experienced and 2 experienced breast imaging radiologists analyzed the US features of the breast masses without and with CAD according to the Breast Imaging Reporting and Data System (BI-RADS) lexicon and categories. We then compared the diagnostic performance between the experienced and less experienced radiologists and analyzed the interobserver agreement among the radiologists. Of the 100 breast masses, 41 (41%) were malignant and 59 (59%) were benign. Compared with the experienced radiologists, the less experienced radiologists had significantly improved negative predictive value (86.7%–94.7% vs 53.3%–76.2%, respectively) and area under receiver operating characteristics curve (0.823–0.839 vs 0.623–0.759, respectively) with CAD assistance (all P < .05). In contrast, experienced radiologists had significantly improved specificity (52.5% and 54.2% vs 66.1% and 66.1%) and positive predictive value (55.6% and 58.5% vs 64.9% and 64.9%, respectively) with CAD assistance (all P < .05). Interobserver variability of US features and final assessment by categories were significantly improved and moderate agreement was seen in the final assessment after CAD combination regardless of the radiologist's experience. CAD is a useful additional diagnostic tool for breast US in all radiologists, with benefits differing depending on the radiologist's level of experience. In this study, CAD improved the interobserver agreement and showed acceptable agreement in the characterization of breast masses.

Jang M, et al. Korean J Radiol. 2014 Jul-Aug;15(4):411-422. https://doi.org/10.3348/kjr.2014.15.4.411

Microcalcifications on mammograms obtained before neoadjuvant chemotherapy (NAC), peritumoral edema on MRI scans obtained after NAC, and clinical-pathologic variables were associated with survival ... Background Little is known regarding findings at imaging associated with survival in patients with luminal breast cancer treated with neoadjuvant chemotherapy (NAC). Purpose To determine the relati...

We investigated addition of strain and shear wave elastography to conventional ultrasonography for the qualitative and quantitative assessment of breast masses; cut-off points were determined for strain ratio, elasticity ratio, and visual score for differentiating between benign and malignant masses. In all, 108 masses from 94 patients were evaluated with strain and shear wave elastography and scored for suspicion of malignancy, visual score, strain ratio, and elasticity ratio. The diagnostic performance between ultrasonography alone and ultrasonography combined with either type of elastography was compared; cut-off points were determined for strain ratio, elasticity ratio, and visual score. Of the 108 masses, 44 were malignant and 64 were benign. The areas under the curves were significantly higher for strain and shear wave elastography-supplemented ultrasonography (0.839 and 0.826, respectively; P = 0.656) than for ultrasonography alone (0.764; P = 0.018 and 0.035, respectively). The diagnostic performances of strain and elasticity ratios were similar when differentiating benign from malignant masses. Cut-off values for strain ratio, elasticity ratio, and visual scores for strain and shear wave elastography were 2.93, 4, 3, and 2, respectively. Both forms of elastography similarly improved the diagnostic performance of conventional ultrasonography in the qualitative and quantitative assessment of breast masses.