The purpose of this study was to develop a computer-aided diagnosis (CAD) system for the classification of malignant and benign masses in the breast using ultrasonography based on a convolutional neural network (CNN), a state-of-the-art deep learning technique. We explored the regions for the correct classification by generating a heat map that presented the important regions used by the CNN for human malignancy/benign classification. Clinical data was obtained from a large-scale clinical trial previously conducted by the Japan Association of Breast and Thyroid Sonology. Images of 1536 breast masses (897 malignant and 639 benign) confirmed by pathological examinations were collected, with each breast mass captured from various angles using an ultrasound (US) imaging probe. We constructed an ensemble network by combining two CNN models (VGG19 and ResNet152) fine-tuned on balanced training data with augmentation and used the mass-level classification method to enable the CNN to classify a given mass using all views. For an independent test set consisting of 154 masses (77 malignant and 77 benign), our network showed outstanding classification performance with a sensitivity of 90.9% (95% confidence interval 84.5-97.3), a specificity of 87.0% (79.5-94.5), and area under the curve (AUC) of 0.951 (0.916-0.987) compared to that of the two CNN models. In addition, our study indicated that the breast masses themselves were not detected by the CNN as important regions for correct mass classification. Collectively, this CNN-based CAD system is expected to assist doctors by improving the diagnosis of breast cancer in clinical practice.

The Japan Association of Breast and Thyroid Sonology (JABTS) proposed, in 2003, a conceptual classification system for non-mass abnormalities to be applied in addition to the conventional concept of masses, to facilitate detecting ductal carcinoma in situ (DCIS) lesions. The aim of this study was to confirm the utility of this system and to clarify the distribution of these findings in DCIS lesions. Data on 705 surgically treated DCIS lesions from 16 institutions in Japan were retrospectively reviewed. All 705 DCIS lesions could be classified according to the JABTS classification system. The most frequent findings were hypo-echoic areas in the mammary gland (48.6%), followed by solid masses (28.0%) and duct abnormalities (10.2%) or mixed masses (8.1%). Distortion (1.3%), clustered microcysts (1.4%) and echogenic foci without a hypo-echoic area (2.5%) were uncommon. These results suggest that the concept of non-mass abnormalities is useful in detecting DCIS lesions.

The use of color Doppler ultrasound (CD) for distinguishing between benign and malignant breast lesions remains controversial. This study (JABTS BC-04 study) was aimed at confirming the usefulness of our CD diagnostic criteria. We evaluated ultrasound images of 1408 solid breast masses from 16 institutions in Japan (malignant: 839, benign: 569). Multivariate analysis indicated that vascularity (amount of blood flow), vascular flow pattern ("surrounding marginal flow" or "penetrating flow") and the incident angle of penetrating flow were significant findings for distinguishing between benign and malignant lesions. However, the sensitivity and specificity of B-mode alone did not improve significantly with CD addition (97.6% ! 97.9%, 38.3% ! 41.5%, respectively). We explored the causes of these negative results and found that age should have been considered when evaluating vascularity. Simulation experiments suggested that specificity is significantly improved when age is taken into consideration (38.3% ! 46.0%, p < 0.001) and we thereby improved our diagnostic criteria.

Purpose: We assessed the in‰uence of the menstrual cycle on background parenchymal enhancement (BPE) of the breast in the early and delayed phases of dynamic magnetic resonance (MR) imaging and the optimal timing of MR imaging of the breast in Japanese women.Material and Methods: We reviewed dynamic MR images of 165 consecutive women with regular menstrual cycles and divided the women into 4 groups by week of the menstrual cycle: 32 in Week One (Days 1 through 4 of the menstrual cycle); 46 in Week 2 (Days 5 through 12); 49 in Week 3 (Days 13 through 20); and 38 in Week 4 (Days 21 through 30). We qualitatively evaluated BPE of the whole breast in the early and delayed phases of MR imaging; categorized enhancement as minimal, mild, moderate, or marked; and calculated the rate at which signal intensity increased (＝SI post-SI pre/SI pre) in regions of interest in from the early and delayed phase to the before contrast administration phase to assess BPE quantitatively.Results: In both the early and delayed dynamic MR phases, BPE was signiˆcantly more extensive and stronger in Week 4 than Week 2 ( Pº0.01). Throughout the menstrual cycle, BPE was signiˆcantly stronger in the delayed phase than in the early phase in both qualitative (Week One, P＝0.0002; Weeks 2 through 4, Pº0.0001) and quantitative (Weeks One through 4, Pº0.0001) assessments.Conclusion: The optimal time to perform dynamic breast MR imaging in premenopausal Japanese women was during Days 5 through 12 of the menstrual cycle.

No abstract