Search

Abstract Background Studies on deep learning (DL)-based models in breast ultrasound (US) remain at the early stage due to a lack of large datasets for training and independent test sets for verification. We aimed to develop a DL model for differentiating benign from malignant breast lesions on US using a large multicenter dataset and explore the model’s ability to assist the radiologists. Methods A total of 14,043 US images from 5012 women were prospectively collected from 32 hospitals. To develop the DL model, the patients from 30 hospitals were randomly divided into a training cohort (n = 4149) and an internal test cohort (n = 466). The remaining 2 hospitals (n = 397) were used as the external test cohorts (ETC). We compared the model with the prospective Breast Imaging Reporting and Data System assessment and five radiologists. We also explored the model’s ability to assist the radiologists using two different methods. Results The model demonstrated excellent diagnostic performance with the ETC, with a high area under the receiver operating characteristic curve (AUC, 0.913), sensitivity (88.84%), specificity (83.77%), and accuracy (86.40%). In the comparison set, the AUC was similar to that of the expert (p = 0.5629) and one experienced radiologist (p = 0.2112) and significantly higher than that of three inexperienced radiologists (p

Background: Graf’s method is currently the most commonly used ultrasound-based technique for the diagnosis of developmental dysplasia of the hip (DDH). However, the efficiency and accuracy of diagnosis are highly affected by the sonographers’ qualification and the time and effort expended, which has a significant intra- and inter-observer variability. Methods: Aiming to minimize the manual intervention in the diagnosis process, we developed a deep learning-based computer-aided framework for the DDH diagnosis, which can perform fully automated standard plane detection and angle measurement for Graf type I and type II hips. The proposed framework is composed of three modules: an anatomical structure detection module, a standard plane scoring module, and an angle measurement module. This framework can be applied to two common clinical scenarios. The first is the static mode, measurement and classification are performed directly based on the given standard plane. The second is the dynamic mode, where a standard plane from ultrasound video is first determined, and measurement and classification are then completed. To the best of our knowledge, our proposed framework is the first CAD method that can automatically perform the entire measurement process of Graf’s method. Results: In our experiments, 1051 US images and 289 US videos of Graf type I and type II hips were used to evaluate the performance of the proposed framework. In static mode, the mean absolute error of α, β angles are 1.71° and 2.40°, and the classification accuracy is 94.71%. In dynamic mode, the mean absolute error of α, β angles are 1.97° and 2.53°, the classification accuracy is 89.51%, and the running speed is 31 fps. Conclusions: Experimental results demonstrate that our fully automated framework can accurately perform standard plane detection and angle measurement of an infant’s hip at a fast speed, showing great potential for clinical application.