Inter-observer agreement on the evaluation of automated breast volume scanner for breast masses

Objective: To explore inter-observer agreement on the evaluation of automated breast volume scanner (ABVS) for breast masses. Methods: A total of 846 breast masses in 630 patients underwent ABVS examinations. The imaging data were independently interpreted by senior and junior radiologists regarding the mass size ([Formula: see text][Formula: see text]cm, [Formula: see text][Formula: see text]cm and total). We assessed inter-observer agreement of BI-RADS lexicons, unique descriptors of ABVS coronal planes. Using BI-RADS 3 or 4a as a cutoff value, the diagnostic performances for 331 masses with pathological results in 253 patients were assessed. Results: The overall agreements were substantial for BI-RADS lexicons ([Formula: see text]–0.779) and the characteristics on the coronal plane of ABVS ([Formula: see text]), except for associated features ([Formula: see text]). However, the overall agreement was moderate for orientation ([Formula: see text]) for the masses [Formula: see text][Formula: see text]cm. The agreements were substantial to be perfect for categories 2, 3, 4, 5 and overall ([Formula: see text]–0.918). However, the agreements were moderate to substantial for categories 4a ([Formula: see text]), 4b ([Formula: see text]), and 4c ([Formula: see text]), except for category 4b of the masses [Formula: see text][Formula: see text]cm ([Formula: see text]). Moreover, for radiologists 1 and 2, there were no significant differences in sensitivity, specificity, accuracy, positive and negative predictive values with BI-RADS 3 or 4a as a cutoff value ([Formula: see text] for all). Conclusion: ABVS is a reliable imaging modality for the assessment of breast masses with good inter-observer agreement.

A Machine Learning Ensemble Based on Radiomics to Predict BI-RADS Category and Reduce the Biopsy Rate of Ultrasound-Detected Suspicious Breast Masses

We developed a machine learning model based on radiomics to predict the BI-RADS category of ultrasound-detected suspicious breast lesions and support medical decision-making towards short-interval follow-up versus tissue sampling. From a retrospective 2015–2019 series of ultrasound-guided core needle biopsies performed by four board-certified breast radiologists using six ultrasound systems from three vendors, we collected 821 images of 834 suspicious breast masses from 819 patients, 404 malignant and 430 benign according to histopathology. A balanced image set of biopsy-proven benign (n = 299) and malignant (n = 299) lesions was used for training and cross-validation of ensembles of machine learning algorithms supervised during learning by histopathological diagnosis as a reference standard. Based on a majority vote (over 80% of the votes to have a valid prediction of benign lesion), an ensemble of support vector machines showed an ability to reduce the biopsy rate of benign lesions by 15% to 18%, always keeping a sensitivity over 94%, when externally tested on 236 images from two image sets: (1) 123 lesions (51 malignant and 72 benign) obtained from two ultrasound systems used for training and from a different one, resulting in a positive predictive value (PPV) of 45.9% (95% confidence interval 36.3–55.7%) versus a radiologists’ PPV of 41.5% (p < 0.005), combined with a 98.0% sensitivity (89.6–99.9%); (2) 113 lesions (54 malignant and 59 benign) obtained from two ultrasound systems from vendors different from those used for training, resulting into a 50.5% PPV (40.4–60.6%) versus a radiologists’ PPV of 47.8% (p < 0.005), combined with a 94.4% sensitivity (84.6–98.8%). Errors in BI-RADS 3 category (i.e., assigned by the model as BI-RADS 4) were 0.8% and 2.7% in the Testing set I and II, respectively. The board-certified breast radiologist accepted the BI-RADS classes assigned by the model in 114 masses (92.7%) and modified the BI-RADS classes of 9 breast masses (7.3%). In six of nine cases, the model performed better than the radiologist did, since it assigned a BI-RADS 3 classification to histopathology-confirmed benign masses that were classified as BI-RADS 4 by the radiologist.

Comparative analysis of proficiencies of various textures and geometric features in breast mass classification using k-nearest neighbor

This paper introduces a comparative analysis of the proficiencies of various textures and geometric features in the diagnosis of breast masses on mammograms. An improved machine learning-based framework was developed for this study. The proposed system was tested using 106 full field digital mammography images from the INbreast dataset, containing a total of 115 breast mass lesions. The proficiencies of individual and various combinations of computed textures and geometric features were investigated by evaluating their contributions towards attaining higher classification accuracies. Four state-of-the-art filter-based feature selection algorithms (Relief-F, Pearson correlation coefficient, neighborhood component analysis, and term variance) were employed to select the top 20 most discriminative features. The Relief-F algorithm outperformed other feature selection algorithms in terms of classification results by reporting 85.2% accuracy, 82.0% sensitivity, and 88.0% specificity. A set of nine most discriminative features were then selected, out of the earlier mentioned 20 features obtained using Relief-F, as a result of further simulations. The classification performances of six state-of-the-art machine learning classifiers, namely k-nearest neighbor (k-NN), support vector machine, decision tree, Naive Bayes, random forest, and ensemble tree, were investigated, and the obtained results revealed that the best classification results (accuracy = 90.4%, sensitivity = 92.0%, specificity = 88.0%) were obtained for the k-NN classifier with the number of neighbors having k = 5 and squared inverse distance weight. The key findings include the identification of the nine most discriminative features, that is, FD26 (Fourier Descriptor), Euler number, solidity, mean, FD14, FD13, periodicity, skewness, and contrast out of a pool of 125 texture and geometric features. The proposed results revealed that the selected nine features can be used for the classification of breast masses in mammograms.

Qualitative and quantitative strain and shear wave elastography paradigm in differentiation of breast lesions

Background Breast cancer is the most common life-threatening cancer in women worldwide. A high number of women are going through biopsy procedures for characterization of breast masses every day and yet 75% of the pathological results prove these masses to be benign. Ultrasound (US) elastography is a non-invasive technique that measures tissue stiffness. It is convenient for differentiating benign from malignant breast tumors. Our study aims to evaluate the role of qualitative ultrasound elastography scoring (ES), quantitative mass strain ratio (SR), and shear wave elasticity ratio (SWER) in differentiation between benign and malignant breast lesions. Results Among 51 female patients with 77 histopathologically proved breast lesions, 57 breast masses were malignant and 20 were benign. All patients were examined by B-mode ultrasound then strain and shear wave elastographic examinations using ultrasound machine (Logiq E9, GE Medical Systems) with 8.5–12 MHz high-frequency probes. Our study showed that ES best cut-off point > 3 with sensitivity, specificity, PPV, NPP, accuracy was 94.7%, 85%, 94.7%, 85%, 90.9%, respectively, and AUC = 0.926 at P < 0.001, mass SR the best cut-off point > 4.6 with sensitivity, specificity, PPV, NPP, accuracy was 96.5%, 80%, 93.2%, 88.9%, 92.2%, respectively, and AUC = 0.860 at P < 0.001, SWER the best cut-off value > 4.9 with sensitivity, specificity, PPV, NPP and accuracy was 91.2%, 80%, 92.9%, 76.2%, 93.5%, respectively, and AUC = 0.890 at P < 0.001. The mean mass strain ratio for malignant lesions is 10.1 ± 3.7 SD and for solid benign lesions 4.7 ± 4.3 SD (p value 0.001). The mean shear wave elasticity ratio for malignant lesions is 10.6 ± 5.4 SD and for benign (solid and cystic) lesions 3.6 ± 4.2 SD. Using ROC curve and Youden index, the difference in diagnostic performance between ES, SR and SWER was not significant in differentiation between benign and malignant breast lesions and also was non-significant difference when comparing them with conventional US alone. Conclusion ES, SR, and SWER have a high diagnostic performance in differentiating malignant from benign breast lesions with no statistically significant difference between them.

Discriminative Localized Sparse Approximations for Mass Characterization in Mammograms

The most common form of cancer among women in both developed and developing countries is breast cancer. The early detection and diagnosis of this disease is significant because it may reduce the number of deaths caused by breast cancer and improve the quality of life of those effected. Computer-aided detection (CADe) and computer-aided diagnosis (CADx) methods have shown promise in recent years for aiding in the human expert reading analysis and improving the accuracy and reproducibility of pathology results. One significant application of CADe and CADx is for breast cancer screening using mammograms. In image processing and machine learning research, relevant results have been produced by sparse analysis methods to represent and recognize imaging patterns. However, application of sparse analysis techniques to the biomedical field is challenging, as the objects of interest may be obscured because of contrast limitations or background tissues, and their appearance may change because of anatomical variability. We introduce methods for label-specific and label-consistent dictionary learning to improve the separation of benign breast masses from malignant breast masses in mammograms. We integrated these approaches into our Spatially Localized Ensemble Sparse Analysis (SLESA) methodology. We performed 10- and 30-fold cross validation (CV) experiments on multiple mammography datasets to measure the classification performance of our methodology and compared it to deep learning models and conventional sparse representation. Results from these experiments show the potential of this methodology for separation of malignant from benign masses as a part of a breast cancer screening workflow.