Objective:To explore the application value of artificial intelligence-assisted diagnosis model based on convolutional neural network (CNN) in the differential diagnosis of benign and malignant breast masses.Methods:A total of 10 490 images of 2 098 patients with breast lumps (including 1 132 cases of benign tumor, 779 cases of malignant tumor, 32 cases of inflammation, 155 cases of adenosis) were collected from January 2016 to January 2018 in Beijing Tiantan Hospital Affiliated to the Capital University of Medical Sciences. They were divided into training set and test set and the auxiliary artificial intelligence diagnosis model was used for training and testing. Two sets of data training models were compared by two-dimensional imaging (2D) and two-dimensional and color Doppler flow imaging (2D-CDFI). The ROC curves of benign breast tumors, malignant tumors, inflammation and adenopathy were analyzed, and the area under the ROC curve (AUC) were calculated.Results:The accuracies of 2D-CDFI ultrasonic model for training group and testing group were significantly improved. ①For benign tumors, the result from training set with 2D image was: sensitivity 92%, specificity 95%, AUC 0.93; the result from training set with 2D-CDFI images was: sensitivity 93%, specificity 95%, AUC 0.93; the result for test set with 2D images was: sensitivity 91%, specificity 96%, AUC 0.94; the result for test set with 2D-CDFI images was: sensitivity 93%, specificity: 94%, AUC 0.94. ② For malignancies, the result for training set with 2D images was: sensitivity 93%, specificity 97%, AUC 0.94; the result for training set with 2D-CDFI images was: sensitivity 93%, specificity 96%, AUC 0.94; the result for test set with 2D images was: sensitivity 93%, specificity 96%, AUC 0.94; the result for test set with 2D-CDFI images was: sensitivity 93%, specificity 96%, AUC 0.94. ③For inflammation, the result for training set with 2D images was: sensitivity 81%, specificity 99%, AUC 0.91; the result for training set with 2D-CDFI images was: sensitivity 86%, specificity 99%, AUC 0.89; the result for test set with 2D images was: sensitivity 100%, specificity 98%, AUC 0.98; the result for test set with 2D-CDFI images was: sensitivity 100%, specificity 99%, AUC 0.96. ④For adenopathy, the result for training set with 2D images was: sensitivity 88%, specificity 97%, AUC 0.94; the result for training set with 2D-CDFI images was: sensitivity 93%, specificity 98%, AUC 0.94; the result for test set with 2D images was: sensitivity 94%, specificity 98%, AUC 0.93; the result for test set with 2D-CDFI images was: sensitivity 88%, specificity 99%, AUC 0.90. Its diastolic accuracy was not affected even if the maximum diameter of the tumor was less than 1 cm.Conclusions:Through the deep learning of artificial intelligence based on CNN for breast masses, it can be more finely classified and the diagnosis rate can be improved. It has potential guiding value for the treatment of breast cancer patients.

Objective:To explore the application value of artificial intelligence (AI) model based on deep learning in breast nodules classification of Breast Imaging Reporting and Data System of ultrasound (BI-RADS-US).Methods:The ultrasound images of 2 426 breast nodules from 1 558 female patients with breast diseases at Beijing Tongren Hospital, Capital Medical University between December 2006 and December 2019 were collected . The image data sets were divided into training (63%), verification (7%), and test (30%) subsets for the construction of AI model. The diagnostic efficiencies of AI model, doctors' arbitration results and doctors' diagnosis with or without AI model assistance were analyzed by using receiver operating characteristic (ROC) curve. The Cohen weighted Kappa statistic was used to compare the consistency of BI-RADS-US classification among 5 ultrasound doctors' diagnosis with or without AI model assistance. And the changes of BI-RADS-US classification were analyzed before and after each doctor adopted AI model assistance.Results:The differences in diagnostic efficiencies of AI model, doctors' arbitration results and doctors' diagnosis with or without AI model assistance were statistically significant (all P > 0.05). The consistency among 5 ultrasound doctors was improved due to AI model assistance and Kappa value was increased from 0.433 (category 3), 0.600 (category 4a), 0.614 (category 4b), 0.570 (category 4c) and 0.495 (category 5) to 0.812, 0.704, 0.823, 0.690 and 0.509 (all P < 0.05), respectively. The upgrade and downgrade of BI-RADS-US classification occurred in 5 doctors after the classification of AI model assistance. Downgrade from category 4 to 3 in benign nodules of 56.6% (47/76) and upgrade from category 4 to 5 in malignant nodules of 69.4% (34/49) were mostly observed. Conclusions:AI-assisted BI-RADS-US classification can effectively improve the consistency of classification among the doctors without reducing the diagnostic efficiency. AI model shows clinical values in reducing unnecessary biopsy of partial benign lesions and increasing diagnostic accuracy of partial malignant lesions through the adjustment of breast nodule classification.

Objective: To explore the application value of the convolutional neural network (CNN)-based artificial intelligence-assisted diagnosis model in the ultrasound differentiation diagnosis of benign and malignant breast nodules. Methods: A total of 7 334 ultrasound images from 1 351 patients with breast nodules including 807 benign cases and 544 malignant cases were retrieved by using the CNN-based artificial intelligence-assisted diagnosis model from Beijing Tongren Hospital of Capital Medical University ultrasound images database between December 2006 and July 2017. The study included training subset (6 162 images), verification subset (555 images), and test subset (617 images), which were performed in the artificial intelligence-assisted diagnosis model. The outcome results of test subset in diagnosis model were compared with the pathological results. The sensitivity, specificity and accuracy of the artificial intelligence-assisted diagnosis model were calculated. Results: After the test of 617 images, the model diagnostic results could be automatically output with a rectangular frame indicating the nodule position, benign and malignant diagnosis, benign and malignant probability values. The diagnosis time was approximately 4 seconds for each nodule. The sensitivity, specificity and accuracy of the diagnostic model in differentiating benign and malignant breast nodules were 84.1%, 95.0% and 91.2% , respectively. Conclusion The CNN-based artificial intelligence-assisted diagnosis model has satisfactory results in the differentiation diagnosis of the benign breast nodules and the malignant ones, which indicating the promising application prospect.