Progress in the Application of Lung Ultrasound for the Evaluation of Neonates with Respiratory Distress Syndrome.

Neonatal respiratory distress syndrome (NRDS) is a common critical disease in neonates. Early diagnosis and timely treatment are crucial. Historically, X-ray imaging was the primary method for diagnosing NRDS. However, this method carries radiation exposure risks, making it unsuitable for dynamic lung condition monitoring. In addition, neonates who are critically ill require bedside imaging, but diagnostic delays are often unavoidable due to equipment transportation and positioning limitations. These challenges have been resolved with the introduction of lung ultrasound (LUS) in neonatal intensive care. The diagnostic efficacy and specificity of LUS for NRDS is superior to that of X-ray. The non-invasive, dynamic, and real-time benefits of LUS also allow for real-time monitoring of lung changes throughout treatment for NRDS, yielding important insights for guiding therapy. In this paper, we examine the ultrasonographic characteristics of NRDS and the recent progress in the application of ultrasound in the diagnosis and treatment of NRDS while aiming to promote wider adoption of this method.

Evaluation of standard breast ultrasonography by adding two-dimensional and three-dimensional shear wave elastography: a prospective, multicenter trial.

OBJECTIVE: To reduce the number of biopsies performed on benign breast lesions categorized as BI-RADS 4-5, we investigated the diagnostic performance of combined two-dimensional and three-dimensional shear wave elastography (2D + 3D SWE) with standard breast ultrasonography (US) for the BI-RADS assessment of breast lesions. METHODS: A total of 897 breast lesions, categorized as BI-RADS 3-5, were subjected to standard breast US and supplemented by 2D SWE only and 2D + 3D SWE analysis. Based on the malignancy rate of less than 2% for BI-RADS 3, lesions assessed by standard breast US were reclassified with SWE assessment. RESULTS: After standard breast US evaluation, 268 (46.1%) participants underwent benign biopsies in BI-RADS 4-5 lesions. By using separated cutoffs for upstaging BI-RADS 3 at 120 kPa and downstaging BI-RADS 4a at 90 kPa in 2D + 3D SWE reclassification, 123 (21.2%) participants underwent benign biopsy, resulting in a 54.1% reduction (123 versus 268). CONCLUSION: Combining 2D + 3D SWE with standard breast US for reclassification of BI-RADS lesions may achieve a reduction in benign biopsies in BI-RADS 4-5 lesions without sacrificing sensitivity unacceptably. CLINICAL RELEVANCE STATEMENT: Combining 2D + 3D SWE with US effectively reduces benign biopsies in breast lesions with categories 4-5, potentially improving diagnostic accuracy of BI-RADS assessment for patients with breast lesions. TRIAL REGISTRATION: ChiCTR1900026556 KEY POINTS: • Reduce benign biopsy is necessary in breast lesions with BI-RADS 4-5 category. • A reduction of 54.1% on benign biopsies in BI-RADS 4-5 lesions was achieved using 2D + 3D SWE reclassification. • Adding 2D + 3D SWE to standard breast US improved the diagnostic performance of BI-RADS assessment on breast lesions: specificity increased from 54 to 79%, and PPV increased from 54 to 71%, with slight loss in sensitivity (97.2% versus 98.7%) and NPV (98.1% versus 98.7%).

Densely connected convolutional networks for ultrasound image based lesion segmentation.

Delineating lesion boundaries play a central role in diagnosing thyroid and breast cancers, making related therapy plans and evaluating therapeutic effects. However, it is often time-consuming and error-prone with limited reproducibility to manually annotate low-quality ultrasound (US) images, given high speckle noises, heterogeneous appearances, ambiguous boundaries etc., especially for nodular lesions with huge intra-class variance. It is hence appreciative but challenging for accurate lesion segmentations from US images in clinical practices. In this study, we propose a new densely connected convolutional network (called MDenseNet) architecture to automatically segment nodular lesions from 2D US images, which is first pre-trained over ImageNet database (called PMDenseNet) and then retrained upon the given US image datasets. Moreover, we also designed a deep MDenseNet with pre-training strategy (PDMDenseNet) for segmentation of thyroid and breast nodules by adding a dense block to increase the depth of our MDenseNet. Extensive experiments demonstrate that the proposed MDenseNet-based method can accurately extract multiple nodular lesions, with even complex shapes, from input thyroid and breast US images. Moreover, additional experiments show that the introduced MDenseNet-based method also outperforms three state-of-the-art convolutional neural networks in terms of accuracy and reproducibility. Meanwhile, promising results in nodular lesion segmentation from thyroid and breast US images illustrate its great potential in many other clinical segmentation tasks.

Comparative diagnostic accuracy of multiparametric magnetic resonance imaging-ultrasound fusion-guided biopsy versus systematic biopsy for clinically significant prostate cancer.

Purpose: To examine the accuracy of transperineal magnetic resonance imaging (MRI)-ultrasound (US) fusion biopsy (FB) in identifying men with prostate cancer (PCa) that has reached a clinically relevant stage. Methods: This investigation enrolled 459 males. In 210 of these patients (FB group), transperineal MRI/US fusion-guided biopsies were performed on the suspicious region, and in 249 others, a systematic biopsy (SB) was performed (SB group). We compared these groups using Gleason scores and rates of cancer detection. Results: PCa cases counted 198/459 (43.1%), including 94/249 (37.8%) in the SB group and 104/210 (49.5%) in the FB group. FB was associated with higher overall diagnostic accuracy relative to SB (88.5% and 72.3%, P = 0.024). FB exhibited greater sensitivity than SB (88.9% and 71.2%, P = 0.025). The area under the curve for FB and SB approaches was 0.837 and 0.737, respectively, such that FB was associated with an 11.9% increase in accuracy as determined based upon these AUC values. Relative to SB, FB was better able to detect high-grade tumors (GS ≥ 7) (78.85% vs. 60.64%, P = 0.025). Conclusion: Transperineal MRI-US fusion targeted biopsy is superior to the systematic one as an approach to diagnosing clinically significant PCa, as it is a viable technical approach to prostate biopsy.

Mammography and Ultrasonography Manifestations of Sclerosing Lymphocytic Lobulitis of the Breast: A Series of Seven Case Reports.

OBJECTIVE: The present study aimed to analyze mammography and ultrasonography (US) manifestations of sclerosing lymphocytic lobulitis (SLL) of the breast. METHODS: A total of 8 pathologically confirmed SLL lesions from seven women (with one patient having bilateral breast lesions) were included in the study. All patients underwent preoperative mammography and US examinations. The findings from both modalities were classified and compared to their corresponding clinical data. RESULTS: Four patients were diagnosed with diabetes mellitus. Mammography results revealed that seven lesions presented as focal asymmetry or asymmetry. Seven lesions were observed as non-mass lesions on US examination. The most commonly observed US lesion features were as follows: seven lesions had focal non-ductal hypoechoic areas (87.5%), seven lesions exhibited posterior shadowing (87.5%), all lesions showed no vascularity or vessels in the rim (100%), no lesion had calcifications (0%), five lesions had an elasticity score of 3 (100%), one lesion showed retraction on the coronal plane (20%), and one lesion displayed a skipping sign on the coronal plane (20%). Based on these US findings, seven lesions (87.5%) were classified as BI-RADS 4. CONCLUSION: The mammography findings for SLL are often nonspecific. However, the US features of SLL typically present as non-mass lesions. The absence of calcification and vascularity and no retraction on the coronal plane inside the lesion may help to differentiate this disease from the conventional forms of breast carcinoma.

RadioLOGIC, a healthcare model for processing electronic health records and decision-making in breast disease.

Digital health data used in diagnostics, patient care, and oncology research continue to accumulate exponentially. Most medical information, and particularly radiology results, are stored in free-text format, and the potential of these data remains untapped. In this study, a radiological repomics-driven model incorporating medical token cognition (RadioLOGIC) is proposed to extract repomics (report omics) features from unstructured electronic health records and to assess human health and predict pathological outcome via transfer learning. The average accuracy and F1-weighted score for the extraction of repomics features using RadioLOGIC are 0.934 and 0.934, respectively, and 0.906 and 0.903 for the prediction of breast imaging-reporting and data system scores. The areas under the receiver operating characteristic curve for the prediction of pathological outcome without and with transfer learning are 0.912 and 0.945, respectively. RadioLOGIC outperforms cohort models in the capability to extract features and also reveals promise for checking clinical diagnoses directly from electronic health records.

Efficiency and impact factors of anatomical intelligence for breast and hand-held ultrasound in lesion detection.

PURPOSE: To investigate the efficiency and impact factors of anatomical intelligence for breast (AI-Breast) and hand-held ultrasound (HHUS) in lesion detection. METHODS: A total of 172 outpatient women were randomly selected, underwent AI-Breast ultrasound (Group AI) once and HHUS twice. HHUS was performed by breast imaging radiologists (Group A) and general radiologists (Group B). For the AI-Breast examination, a trained technician performed the whole-breast scan and data acquisition, while other general radiologists performed image interpretation. The examination time and lesion detection rate were recorded. The impact factors for breast lesion detection, including breast cup size, number of lesions, and benign or malignant lesions were analyzed. RESULTS: The detection rates of Group AI, A, and B were 92.8 ± 17.0%, 95.0 ± 13.6%, and 85.0 ± 22.9%, respectively. Comparable lesion detection rates were observed in Group AI and Group A (P > 0.05), but a significantly lower lesion detection rate was observed in Group B compared to the other two (both P < 0.05). Regarding missed diagnosis rates of malignant lesions, comparable performance was observed in Group AI, Group A, and Group B (8% vs. 4% vs. 14%, all P > 0.05). Scan times of Groups AI, A, and B were 262.15 ± 40.4 s, 237.5 ± 110.3 s, 281.2 ± 86.1 s, respectively. The scan time of Group AI was significantly higher than Group A (P < 0.01), but was slightly lower than Group B (P > 0.05). We found a strong linear correlation between scan time and cup size in Group AI (r = 0.745). No impacts of cup size and number of lesions were found on the lesion detection rate in Group AI (P > 0.05). CONCLUSIONS: With the assist of AI-Breast system, the lesion detection rate of AI-Breast ultrasound was comparable to that of a breast imaging radiologist and superior to that of the general radiologist. AI-Breast ultrasound may be used as a potential approach for breast lesions surveillance.

Multi-modal artificial intelligence for the combination of automated 3D breast ultrasound and mammograms in a population of women with predominantly dense breasts.

OBJECTIVES: To assess the stand-alone and combined performance of artificial intelligence (AI) detection systems for digital mammography (DM) and automated 3D breast ultrasound (ABUS) in detecting breast cancer in women with dense breasts. METHODS: 430 paired cases of DM and ABUS examinations from a Asian population with dense breasts were retrospectively collected. All cases were analyzed by two AI systems, one for DM exams and one for ABUS exams. A selected subset (n = 152) was read by four radiologists. The performance of AI systems was based on analysis of the area under the receiver operating characteristic curve (AUC). The maximum Youden's index and its associated sensitivity and specificity were also reported for each AI systems. Detection performance of human readers in the subcohort of the reader study was measured in terms of sensitivity and specificity. RESULTS: The performance of the AI systems in a multi-modal setting was significantly better when the weights of AI-DM and AI-ABUS were 0.25 and 0.75, respectively, than each system individually in a single-modal setting (AUC-AI-Multimodal = 0.865; AUC-AI-DM = 0.832, p = 0.026; AUC-AI-ABUS = 0.841, p = 0.041). The maximum Youden's index for AI-Multimodal was 0.707 (sensitivity = 79.4%, specificity = 91.2%). In the subcohort that underwent human reading, the panel of four readers achieved a sensitivity of 93.2% and specificity of 32.7%. AI-multimodal achieves superior or equal sensitivity as single human readers at the same specificity operating points on the ROC curve. CONCLUSION: Multimodal (ABUS + DM) AI systems for detecting breast cancer in women with dense breasts are a potential solution for breast screening in radiologist-scarce regions.

Value of multimodal imaging in the diagnosis of breast sclerosing adenosis associated with malignant lesions.

AIM: To explore the diagnostic value of multimodal imaging techniques, including automatic breast volume scanner (ABVS), mammography (MG), and magnetic resonance (MRI) in breast sclerosing adenosis (SA) associated with malignant lesions. METHODS: From January 2018 to October 2020, 76 patients (88 lesions) with pathologically confirmed as SA associated with malignant or benign lesions were retrospective analyzed. All patients completed ABVS examination, 58 patients (67 lesions) with MG and 50 patients (62 lesions) with MRI were also completed before biopsy or surgical excision, of which, six patients (eight lesions) diagnosed as Breast Imaging Reporting and Data System (BI-RADS) category 3 by all imaging examinations underwent surgical excision without biopsy, other 70 patients (80 lesions) with BI-RADS category 4 or above by any imaging examination completed biopsy, including 65 patients (75 lesions) were further surgical excised and the other five patients (five lesions) were just followed up. All lesions were retrospectively described and classified, and were divided into benign group and malignant group according to their pathological results. Image features of different examination methods between the two groups were compared and analyzed. A ROC curve was established using the sensitivity of BI-RADS categories to predict malignant lesions in different imaging techniques as the ordinate and 1-specificity as the abscissa. RESULTS: 88 lesions including 26 purely SA and 45 SA associated with benign lesions were classified as benign group, and the remaining 17 SA associated with malignant lesions were classified as malignant group. On ABVS, 40 mass lesions, their heterogeneous echo, not circumscribed margin and coronal convergence signs were statistically significant for malignant lesions (p < .05), but the remain 48 nonmass lesions lack specific sonographic features. On MG, 12 showed negative results, 55 showed with microcalcification, mass, structural distortion, and asymmetric density shadow, of which 11 lesions had the above two signs at the same time, but only microcalcification had statistical difference between the two groups. 35 mass enhanced lesions and 27 nonmass enhanced lesions on MRI, but there were no significant difference between their pathological results. Time signal intensity curves showed no differences, but ADC value <1.10 × 10-3  mm2 /s is more significant in malignant lesions (p < .05). The area under the ROC curve (AUC) of BI-RADS classification of ABVS, MG, and MRI in the diagnosis of malignant lesions were 0.611, 0.474, and 0.751, respectively, and the AUC of the combined diagnosis of the three was 0.761. CONCLUSION: Mass lesions with heterogeneous echo, not circumscribed margin and coronal convergence sign on ABVS, microcalcification on MG and the ADC value <1.10 × 10-3  mm2 /s on MRI are significant signs for SA associated with malignant lesions. The combined diagnosis of the three methods was the highest, and the following were MRI, ABVS, and MG. Therefore, be cognizant of significant characteristics in SA associated with malignancy showed in different imaging examinations can improve the preoperative evaluation of SA and better provide basis for subsequent clinical decision-making.

Predicting papillary thyroid carcinoma cervical lymph node metastases: an algorithm using the American College of Radiology Thyroid Imaging, Reporting and Data System.

BACKGROUND: It is important to predict lymph node metastasis (LNM) in papillary thyroid carcinoma (PTC) preoperatively; however, the relationship between the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS) score and cervical LNM remains unclear. PURPOSE: To evaluate the association between the ACR TI-RADS score and cervical LNM in patients with PTC. MATERIAL AND METHODS: This retrospective study consisted of 474 patients with 548 PTCs. Cervical LNM including central LNM (CLNM) and lateral LNM (LLNM) were confirmed by pathology. Univariate and multivariate analyses were performed to investigate the risk factors of CLNM and LLNM. RESULTS: Multivariate logistic regression analyses indicated that younger age and multifocality were risk factors for CLNM in PTCs with TR5. In addition, younger age, larger tumor size, and Hashimoto's thyroiditis (HT) were risk factors for LLNM in PTCs ≥ 10 mm with TR5. In PTCs with TR4, ACR TI-RADS scores 5-6 conferred risks for LNM. In PTCs ≥ 10 mm with TR5, ACR TI-RADS scores ≥9 were risk factors for LLNM. CONCLUSION: A higher ACR TI-RADS score is a predictor for cervical LNM in PTCs with TR4 and PTCs ≥ 10 mm with TR5.

Predicting the molecular subtypes of breast cancer using nomograms based on three-dimensional ultrasonography characteristics.

Background: Molecular subtyping of breast cancer is commonly doneforindividualzed cancer management because it may determines prognosis and treatment. Therefore, preoperativelyidentifying different molecular subtypes of breast cancery can be significant in clinical practice.Thisretrospective study aimed to investigate characteristic three-dimensional ultrasonographic imaging parameters of breast cancer that are associated with the molecular subtypes and establish nomograms to predict the molecular subtypes of breast cancers. Methods: A total of 309 patients diagnosed with breast cancer between January 2017and December 2019 were enrolled. Sonographic features were compared between the different molecular subtypes. A multinomial logistic regression model was developed, and nomograms were constructed based on this model. Results: The performance of the nomograms was evaluated in terms of discrimination and calibration.Variables such as maximum diameter, irregular shape, non-parallel growth, heterogeneous internal echo, enhanced posterior echo, lymph node metastasis, retraction phenomenon, calcification, and elasticity score were entered into the multinomial model.Three nomograms were constructed to visualize the final model. The probabilities of the different molecular subtypes could be calculated based on these nomograms. Based on the receiver operating characteristic curves of the model, the macro-and micro-areaunder the curve (AUC) were0.744, and 0.787. The AUC was 0.759, 0.683, 0.747 and 0.785 for luminal A(LA), luminal B(LB), human epidermal growth factor receptor 2-positive(HER2), and triple-negative(TN), respectively.The nomograms for the LA, HER2, and TN subtypes provided good calibration. Conclusions: Sonographic features such as calcification and posterior acoustic features were significantly associated with the molecular subtype of breast cancer. The presence of the retraction phenomenon was the most important predictor for the LA subtype. Nomograms to predict the molecular subtype were established, and the calibration curves and receiver operating characteristic curves proved that the models had good performance.

Nodular fasciitis of the breast: the report of three cases.

BACKGROUND: Nodular Fasciitis is a benign fibroblastic proliferation in soft tissues, which mostly occurs in the upper extremities, trunk, head and neck region. It is rarely reported to occur in the breast. CASE PRESENTATION: Herein, we present sonograms of nodular fasciitis in the breast at different durations in three cases. In Case 1, we provided the longest follow-up time in all literatures. In Case 2 and Case 3, we provided the automated breast ultrasound finding of breast nodular fasciitis for the first time. CONCLUSION: Nodular Fasciitis shows clinical features and ultrasonography findings are similar to those of breast cancer. For superficially located breast lesions with a single and rapid growth, nodular fasciitis may be considered in the differential diagnosis of benign entities resembling malignant tumors on breast imaging.

Prospective Evaluation of Ultrasound in a Novel Position with MRI Virtual Navigation for MRI-Detected Only Breast Lesions: A Pilot Study of a More Efficient and Economical Method.

The aim of this study was to evaluate the clinical utility of ultrasound (US) with magnetic resonance imaging (MRI) virtual navigation in a novel prone position for MRI-detected incidental breast lesions. Between June 2016 and June 2020, 30 consecutive patients with 33 additional Breast Imaging Reporting and Data System (BI-RADS) category 4 or 5 lesions that were detected on MRI but occult on second-look US were enrolled in the study. All suspicious lesions were located in real-time US using MRI virtual navigation in the prone position and then followed by US-guided biopsy or surgical excision. Pathological results were taken as the standard of reference. The detection rate of US with MRI virtual navigation was calculated. The MRI features and pathological types of these lesions were analyzed. A total of 31 lesions were successfully located with real-time US with MRI virtual navigation and then US-guided biopsy or localization, and the detection rate was 93.9% (31/33). Twenty-seven (87.1%, 27/31) proved to be benign lesions and four (12.9%, 4/31) were malignant lesions at pathology. Of the 33 MRI-detected lesions, 31 (93.9%, 31/33) were non-mass enhancements and two (6.1%, 2/33) were masses. This study showed that real-time US with prone MRI virtual navigation is a novel efficient and economical method to improve the detection and US-guided biopsy rate of breast lesions that are detected solely on MRI.

Transfer learning for automatic joint segmentation of thyroid and breast lesions from ultrasound images.

PURPOSE: It plays a significant role to accurately and automatically segment lesions from ultrasound (US) images in clinical application. Nevertheless, it is extremely challenging because distinct components of heterogeneous lesions are similar to background in US images. In our study, a transfer learning-based method is developed for full-automatic joint segmentation of nodular lesions. METHODS: Transfer learning is a widely used method to build high performing computer vision models. Our transfer learning model is a novel type of densely connected convolutional network (SDenseNet). Specifically, we pre-train SDenseNet based on ImageNet dataset. Then our SDenseNet is designed as a multi-channel model (denoted Mul-DenseNet) for automatically jointly segmenting lesions. As comparison, our SDenseNet using different transfer learning is applied to segmenting nodules, respectively. In our study, we find that more datasets for pre-training and multiple pre-training do not always work in segmentation of nodules, and the performance of transfer learning depends on a judicious choice of dataset and characteristics of targets. RESULTS: Experimental results illustrate a significant performance of the Mul-DenseNet compared to that of other methods in the study. Specially, for thyroid nodule segmentation, overlap metric (OM), dice ratio (DR), true-positive rate (TPR), false-positive rate (FPR) and modified Hausdorff distance (MHD) are [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text] and [Formula: see text] mm, respectively; for breast nodule segmentation, OM, DR, TPR, FPR and MHD are [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text] and [Formula: see text] mm, respectively. CONCLUSIONS: The experimental results illustrate our transfer learning models are very effective in segmentation of lesions, which also demonstrate that it is potential of our proposed Mul-DenseNet model in clinical applications. This model can reduce heavy workload of the physicians so that it can avoid misdiagnosis cases due to excessive fatigue. Moreover, it is easy and reproducible to detect lesions without medical expertise.

The hidden breast lesions: A case report of bilateral breast cancer.

Bilateral breast cancer (BBC) is rare and is associated with an unfavorable prognosis. Consequently it is crucial to improve diagnostic performance of breast cancer in the clinical setting. We report a case of BBC in a 66-year-old woman and describe the imaging findings, including mammography, hand-held ultrasound, automated breast ultrasound, anatomical intelligence for breast ultrasound (AI-breast), and magnetic resonance imaging. Only AI-breast ultrasound successfully located the two tumors, while other imaging examinations failed to detect the tumor in the right breast.

The diagnostic performance of automated versus handheld breast ultrasound and mammography in symptomatic outpatient women: a multicenter, cross-sectional study in China.

OBJECTIVES: The purpose of this study was to evaluate the diagnostic performance of automated breast ultrasound (ABUS) for breast cancer by comparing it to handheld ultrasound (HHUS) and mammography (MG). METHODS: A multicenter cross-sectional study was conducted between February 2016 and March 2017 in five tertiary hospitals in China, and 1922 women aged 30-69 years old were recruited. Women aged 30-39 years (group A) underwent ABUS and HHUS, and women aged 40-69 (group B) underwent additional MG. Images were interpreted using the Breast Imaging Reporting and Data System (BI-RADS). All BI-RADS 4 and 5 cases were confirmed pathologically. Sensitivities and specificities of all modalities were compared. RESULTS: There were 83 cancers in 677 women in group A and 321 cancers in 1245 women in group B. In the whole study population, the sensitivities of ABUS and HHUS were 92.8% (375/404) and 96.3% (389/404), and the specificities were 93.0% (1411/1518) and 89.6% (1360/1518), respectively. ABUS had a significantly higher specificity to HHUS (p < 0.01), while HHUS had higher sensitivity (p = 0.01). In group B, the sensitivities of ABUS, HHUS, and MG were 93.5% (300/321), 96.6% (310/321), and 87.9% (282/321). The specificities were 93.0% (859/924), 89.9% (831/924), and 91.6% (846/924). ABUS had significantly higher sensitivity (p = 0.02) and comparable specificity compared with MG (p = 0.14). CONCLUSION: ABUS increased sensitivity and had similar specificity compared with mammography in the diagnosis of breast cancer. Additionally, ABUS has comparable performance to HHUS in women aged 30-69 years old. ABUS or HHUS is a suitable modality for breast cancer diagnosis. KEY POINTS: • In breast cancer diagnosis settings, automated breast ultrasound has a higher cancer detection rate, sensitivity, and specificity than mammography, especially in women with dense breasts. • Compared with handheld ultrasound, automated breast ultrasound has higher specificity, lower sensitivity, and comparable diagnostic performance. • Automated breast ultrasound is a suitable modality for breast cancer diagnosis, and may have a potential indication for its further use in the breast cancer early detection.

Diagnostic performance of automated breast ultrasound and handheld ultrasound in women with dense breasts.

PURPOSE: As an adjunct to mammography, ultrasound can improve the detection of breast cancer in women with dense breasts. We aimed to evaluate the diagnostic performance of automated breast ultrasound system (ABUS) and handheld ultrasound (HHUS) in Chinese women with dense breasts, both in combination with mammography and separately. METHODS: This is a cross-sectional multicenter clinical research study. Nine hundred and thirty-seven women with dense breasts underwent ABUS, HHUS, and mammography at one of five tertiary-care hospitals. The diagnostic performance of ABUS and HHUS was evaluated in combination with mammography, or separately in women with mammography-negative dense breasts. The agreement between ABUS and HHUS in breast cancer detection was also assessed. RESULTS: The sensitivity of the combination of ABUS or HHUS with mammography was 99.1% (219/221), and the specificities were 86.9% (622/716) and 84.9% (608/716), respectively. The area under the curve was 0.93 for ABUS combined with mammography and 0.92 for that of HHUS combined with mammography. Statistically significant agreement between ABUS and HHUS in breast cancer detection was observed (percent agreement = 0.94, κ = 0.85). The incremental cancer detection rate in mammography-negative dense breasts was 42.8 per 1000 ultrasound examinations. CONCLUSIONS: Both ABUS and HHUS as adjuncts to mammography can significantly improve the breast cancer detection rate in women with dense breasts, and there is a strong correlation between them. Given the high prevalence of dense breasts and the multiple advantages of ABUS over HHUS, such as less operator dependence and reproducibility, ABUS showed great potential for use in breast cancer early detection, especially in resource-limited areas.

Lymph Node Metastasis Prediction from Primary Breast Cancer US Images Using Deep Learning.

Background Deep learning (DL) algorithms are gaining extensive attention for their excellent performance in image recognition tasks. DL models can automatically make a quantitative assessment of complex medical image characteristics and achieve increased accuracy in diagnosis with higher efficiency. Purpose To determine the feasibility of using a DL approach to predict clinically negative axillary lymph node metastasis from US images in patients with primary breast cancer. Materials and Methods A data set of US images in patients with primary breast cancer with clinically negative axillary lymph nodes from Tongji Hospital (974 imaging studies from 2016 to 2018, 756 patients) and an independent test set from Hubei Cancer Hospital (81 imaging studies from 2018 to 2019, 78 patients) were collected. Axillary lymph node status was confirmed with pathologic examination. Three different convolutional neural networks (CNNs) of Inception V3, Inception-ResNet V2, and ResNet-101 architectures were trained on 90% of the Tongji Hospital data set and tested on the remaining 10%, as well as on the independent test set. The performance of the models was compared with that of five radiologists. The models' performance was analyzed in terms of accuracy, sensitivity, specificity, receiver operating characteristic curves, areas under the receiver operating characteristic curve (AUCs), and heat maps. Results The best-performing CNN model, Inception V3, achieved an AUC of 0.89 (95% confidence interval [CI]: 0.83, 0.95) in the prediction of the final clinical diagnosis of axillary lymph node metastasis in the independent test set. The model achieved 85% sensitivity (35 of 41 images; 95% CI: 70%, 94%) and 73% specificity (29 of 40 images; 95% CI: 56%, 85%), and the radiologists achieved 73% sensitivity (30 of 41 images; 95% CI: 57%, 85%; P = .17) and 63% specificity (25 of 40 images; 95% CI: 46%, 77%; P = .34). Conclusion Using US images from patients with primary breast cancer, deep learning models can effectively predict clinically negative axillary lymph node metastasis. Artificial intelligence may provide an early diagnostic strategy for lymph node metastasis in patients with breast cancer with clinically negative lymph nodes. Published under a CC BY 4.0 license. Online supplemental material is available for this article. See also the editorial by Bae in this issue.

Diagnostic Performance Using Automated Breast Ultrasound System for Breast Cancer in Chinese Women Aged 40 Years or Older: A Comparative Study.

The purpose of this study was to investigate the diagnostic performance of the automated breast ultrasound system (ABUS) compared with hand-held ultrasonography (HHUS) and mammography (MG) for breast cancer in women aged 40 y or older. A total of 594 breasts in 385 patients were enrolled in the study. HHUS, ABUS and MG exams were performed for these patients. Follow-up and pathologic findings were used as the reference standard. Based on the reference standard, 519 units were benign or normal and 75 were malignant. The sensitivity, specificity, accuracy and Youden index were 97.33%, 89.79%, 90.74% and 0.87 for HHUS; 90.67%, 92.49%, 92.26% and 0.83 for ABUS; 84.00%, 92.87%, 91.75% and 0.77 for MG, respectively. The specificity of ABUS was significantly superior to that of HHUS (p = 0.024). The area under the receiver operating characteristic curve was 0.936 for HHUS, which was the highest, followed by 0.916 for ABUS and 0.884 for MG. However, the difference was not statistically significant (p > 0.05). In conclusion, the diagnostic performance of ABUS for breast cancer was equivalent to HHUS and MG and potentially can be used as an alternative method for breast cancer diagnosis.