Objective To explore the value of four deep learning(DL)models for classifying 7 cardiac ultrasound views of normal fetus.Methods Two hundred normal fetuses who received fetal cardiac ultrasound examinations in 18 to 24+6 weeks of gestation were retrospectively included and divided into training set(n=140)and test set(n=60)at a ratio of 7:3.Two-dimensional ultrasound images were collected,including three-vessel and trachea(3VT)view,apical four-chamber(A4C)view,aortic arch long-axis view,bicaval view,left ventricular outflow tract(LVOT)view,three-vessel(3V)view and right ventricular outflow tract(RVOT)view.After image preprocessing,image features were extracted,and then 4 different DL models were constructed for classifying normal fetal cardiac ultrasound views,i.e.Vision Transformer(ViT),Data-efficient Image Transformer(DeiT),Vision-long short term memory(ViL)and Multi-axis Vision Transformer(MaxViT).The classification performance of each model in test set was assessed with accuracy,precision,recall and F1 score.Gradient-weighted class activation mapping(Grad-CAM)was used to obtain heatmaps for visualizing regions with the most distinctive features on ultrasound images.Results All ViT,DeiT,ViL and MaxViT had excellent performance in classifying normal fetal cardiac ultrasound views in test set,among which MaxViT was the optimal one,with accuracy,precision,recall and F1 score of 98.93％,98.93％,98.95％and 98.93％,respectively.Grad-CAM visualization results indicated that for classification of 7 cardiac ultrasound views of normal fetus using DL models,the heart and vessels present as the deepest red color,indicating the greatest contribution to the classification,also got the highest attention these models.Conclusion The obtained 4 DL models,especially MaxViT,had good capability for classifying normal fetal cardiac ultrasound views,with the interpretability of classifying results validated by Grad-CAM.

Objective:To evaluate the value of the proposed interpretable denoising deep learning model-grouped sharing convolutional attention-visual transformer（GSCAViT）for classifying normal fetal echocardiograms.Methods:A total of 2 501 images from 358 fetuses who underwent cardiac ultrasound examinations at Xiangya Second Hospital of Central South University from January to November 2024 were retrospectively analyzed. GSCAViT was constructed based on fetal echocardiograms from the three-vessel and trachea view，apical four-chamber view，long-axis view of the aortic arch，bicaval view，left ventricular outflow tract view，three-vessel view and right ventricular outflow tract view were compared with both baseline and improved models in the validation set to assess the performance of the classification echocardiography in terms of accuracy，precision，recall and F1-score. Its generalizability across test sets was assessed using the area under the ROC curve（AUC），sensitivity，specificity and F1-score. The impact of image features was interpreted using SHapley Additive exPlanations（SHAP）.The effectiveness of the GSCA module was compared through visual analysis，image parameter metrics and classification performance.Results:The GSCAViT model achieved classification performance for fetal echocardiograms second only to MaxViT，with an accuracy of 97.1%，precision of 97.1%，recall of 97.0%，and an F1-score of 97.0%. In the E10，E20 and E8 test sets，the AUCs of GSCAViT for the prediction of 7 types of fetal echocardiograms were 0.994，0.928 and 0.932，the sensitivities were 99.4%，81.3% and 72.9%，the specificities were 99.7%，96.8% and 94.8%，the F1-scores were 99.4%，81.3% and 67.6%，respectively. SHAP visualization indicated that the model could identify key structural features within the images. Images processed by the denoising-guided group-sharing convolutional attention module best captured and enhanced important regional features，achieving the highest contrast-to-noise ratio，peak signal-to-noise ratio and optimal classification performance，which demonstrated the module's effectiveness.Conclusions:The proposed GSCAViT model exhibits superior performance in classifying seven types of normal fetal echocardiograms compared to the baseline and some improved models. Furthermore，SHAP visualization enhances the interpretability of the classification results，and visual comparisons，image parameter analyses，as well as classification performance metrics confirming the effectiveness of the denoising-guided group-sharing convolutional attention module in the visual transformer model.

Objective To explore the value of four deep learning(DL)models for classifying 7 cardiac ultrasound views of normal fetus.Methods Two hundred normal fetuses who received fetal cardiac ultrasound examinations in 18 to 24+6 weeks of gestation were retrospectively included and divided into training set(n=140)and test set(n=60)at a ratio of 7:3.Two-dimensional ultrasound images were collected,including three-vessel and trachea(3VT)view,apical four-chamber(A4C)view,aortic arch long-axis view,bicaval view,left ventricular outflow tract(LVOT)view,three-vessel(3V)view and right ventricular outflow tract(RVOT)view.After image preprocessing,image features were extracted,and then 4 different DL models were constructed for classifying normal fetal cardiac ultrasound views,i.e.Vision Transformer(ViT),Data-efficient Image Transformer(DeiT),Vision-long short term memory(ViL)and Multi-axis Vision Transformer(MaxViT).The classification performance of each model in test set was assessed with accuracy,precision,recall and F1 score.Gradient-weighted class activation mapping(Grad-CAM)was used to obtain heatmaps for visualizing regions with the most distinctive features on ultrasound images.Results All ViT,DeiT,ViL and MaxViT had excellent performance in classifying normal fetal cardiac ultrasound views in test set,among which MaxViT was the optimal one,with accuracy,precision,recall and F1 score of 98.93％,98.93％,98.95％and 98.93％,respectively.Grad-CAM visualization results indicated that for classification of 7 cardiac ultrasound views of normal fetus using DL models,the heart and vessels present as the deepest red color,indicating the greatest contribution to the classification,also got the highest attention these models.Conclusion The obtained 4 DL models,especially MaxViT,had good capability for classifying normal fetal cardiac ultrasound views,with the interpretability of classifying results validated by Grad-CAM.

Objective:To evaluate the value of the proposed interpretable denoising deep learning model-grouped sharing convolutional attention-visual transformer（GSCAViT）for classifying normal fetal echocardiograms.Methods:A total of 2 501 images from 358 fetuses who underwent cardiac ultrasound examinations at Xiangya Second Hospital of Central South University from January to November 2024 were retrospectively analyzed. GSCAViT was constructed based on fetal echocardiograms from the three-vessel and trachea view，apical four-chamber view，long-axis view of the aortic arch，bicaval view，left ventricular outflow tract view，three-vessel view and right ventricular outflow tract view were compared with both baseline and improved models in the validation set to assess the performance of the classification echocardiography in terms of accuracy，precision，recall and F1-score. Its generalizability across test sets was assessed using the area under the ROC curve（AUC），sensitivity，specificity and F1-score. The impact of image features was interpreted using SHapley Additive exPlanations（SHAP）.The effectiveness of the GSCA module was compared through visual analysis，image parameter metrics and classification performance.Results:The GSCAViT model achieved classification performance for fetal echocardiograms second only to MaxViT，with an accuracy of 97.1%，precision of 97.1%，recall of 97.0%，and an F1-score of 97.0%. In the E10，E20 and E8 test sets，the AUCs of GSCAViT for the prediction of 7 types of fetal echocardiograms were 0.994，0.928 and 0.932，the sensitivities were 99.4%，81.3% and 72.9%，the specificities were 99.7%，96.8% and 94.8%，the F1-scores were 99.4%，81.3% and 67.6%，respectively. SHAP visualization indicated that the model could identify key structural features within the images. Images processed by the denoising-guided group-sharing convolutional attention module best captured and enhanced important regional features，achieving the highest contrast-to-noise ratio，peak signal-to-noise ratio and optimal classification performance，which demonstrated the module's effectiveness.Conclusions:The proposed GSCAViT model exhibits superior performance in classifying seven types of normal fetal echocardiograms compared to the baseline and some improved models. Furthermore，SHAP visualization enhances the interpretability of the classification results，and visual comparisons，image parameter analyses，as well as classification performance metrics confirming the effectiveness of the denoising-guided group-sharing convolutional attention module in the visual transformer model.

Objective:To study the association between preoperative hemoglobin amount and incidence of lower limb deep vein thrombosis (DVT) in patients with lower limb fracture.Methods:A retrospective study was performed of the 2, 482 patients with lower limb fracture who had been treated at Department of Orthopaedics Trauma, Honghui Hospital Affiliated to Xi'an Jiaotong University from July 2014 to August 2019. They were 1, 174 males and 1, 308 females with an age of (60.6±19.3) years. Recorded were the patients' age, gender, injury time, hemoglobin amount, D-dimer measurement, combined medical conditions, time and results of ultrasound vein examination on both lower extremities. According to the ultrasound results, the patients were divided into a thrombosis group and a thrombosis-free group. The 2 groups were compared in hemoglobin amount. Logistic regression was used to analyze the relationship between preoperative hemoglobin amount and incidence of lower limb DVT. The patients were divided into 5 groups according to the quintile of hemoglobin amount; the incidences of thrombosis were compared between the 5 groups.Results:The total incidence of DVT in this cohort was 29.53%(733/2, 482). The hemoglobin amount in the thrombosis group was (116.57±19.24) g/L, significantly lower than that in the thrombosis-free group (124.76±19.79) g/L ( P<0.05). The preoperative hemoglobin amount was a risk factor for incidence of DVT after a lower limb fracture ( OR=0.985, 95% CI: 0.980 to 0.990, P<0.001). As the quintile level of hemoglobin increased, the incidence of DVT showed a downward trend. In comparison of the group with the highest DVT incidence (40.58%) and the group with the lowest DVT incidence (17.27%), the risk increased by 2.386 times (95% CI: 1.718 to 3.315). Conclusions:The preoperative hemoglobin amount can affect the DVT incidence after a lower limb fracture, and a low hemoglobin amount may more likely lead to lower limb DVT.

Objective:To investigate the differences in incidence of deep vein thrombosis (DVT) after closed fracture of lower extremity between patients with different blood types ABO.Methods:A retrospective study was conducted in the 1, 951 patients who had been admitted to Department of Orthopaedics Trauma, Honghui Hospital Affiliated to Xi'an Jiaotong University for lower extremity fractures from August 2014 to June 2018. They were 924 males and 1,027 females with a mean age of 63 (46, 78) years (range, from 16 to 102 years). Of them, 572 were type O, 564 type A, 609 type B and 206 type AB. Venous ultrasonography was performed on both lower extremities within 12 hours after admission. The incidences of DVT after fracture were compared between different blood types in all the patients, patients with proximal fracture of the knee, peri-knee fracture and distal fracture of the knee, and patients<60 years old and ≥60 years old.Results:The incidences of DVT were, respectively, 26.75% (153/572), 28.72% (162/564), 34.32% (209/609) and 29.61% (61/206) for patients with blood type O, type A, type B and type AB. The DVT incidence for type B was significantly higher than that for type O ( P< 0.008). The incidences of DVT were, respectively, 28.74% (98/341), 28.99% (100/345), 39.45% (144/365) and 30.97% (35/113) for blood type O, type A, type B and type AB in the patients with proximal fracture of the knee. The DVT incidence for blood type B was significantly higher than those for blood type O and blood type A ( P< 0.008). There were no significant differences in the DVT incidence between different blood types ABO in the patients with peri-knee fracture, distal fracture of the knee,<60 years old or ≥60 years old( P>0.05). The incidences of DVT were, respectively, 30.99% (97/313), 33.33% (108/324), 45.22% (156/345), 34.74% (33/95) for blood type O, type A, type B and type AB in the patients ≥60 years old. The DVT incidence for blood type B was significantly higher than those for blood type O and blood type A ( P< 0.008). Conclusions:The incidence of DVT varied with different blood types ABO after lower extremity fracture. The highest DVT incidence was found in patients with blood type B. The impact of blood type on the DVT incidence after lower extremity fracture was mainly observed in the patients with proximal fracture of the knee or an age of ≥ 60 years old.

Objective:To study the association between histopathological features and HER2 overexpression/amplification in breast cancers using deep learning algorithms.Methods:A total of 345 HE-stained slides of breast cancer from 2012 to 2018 were collected at the China-Japan Friendship Hospital, Beijing, China. All samples had accurate diagnosis results of HER2 which were classified into one of the 4 HER2 expression levels (0, 1+, 2+, 3+). After digitalization, 204 slides were used for weakly supervised model training, and 141 used for model testing. In the training process, the regions of interest were extracted through cancer detected model and then input to the weakly supervised classification model to tune the model parameters. In the testing phase, we compared performance of the single- and double-threshold strategies to assess the role of the double-threshold strategy in clinical practice.Results:Under the single-threshold strategy, the deep learning model had a sensitivity of 81.6% and a specificity of 42.1%, with the AUC of 0.67 [95% confidence intervals (0.560,0.778)]. Using the double-threshold strategy, the model achieved a sensitivity of 96.3% and a specificity of 89.5%.Conclusions:Using HE-stained histopathological slides alone, the deep learning technology could predict the HER2 status using breast cancer slides, with a satisfactory accuracy. Based on the double-threshold strategy, a large number of samples could be screened with high sensitivity and specificity.

Objective:To investigate the risk factors of deep venous thrombosis in patients before hip arthroplasty, and to explore the possible causes of postoperative thrombosis.Methods:The clinical data of 361 patients with hip arthroplasty treated in the Department of Orthopaedic Trauma of Xi′an Honghui Hospital from September 2015 to December 2019 were studied retrospectively, including 102 males and 259 females, aged 65 to 94 years, and the average age was 72.25 years old. All fracture patients were given subcutaneous injection of low molecular weight heparin calcium to prevent lower extremity thrombosis. The deep veins of both lower extremities were examined before and after operation. The general data of the two groups of patients were collected and recorded, including age, sex, whether complicated with medical diseases (essential hypertension, type 2 diabetes, coronary heart disease), serological indexes, time from injury to admission, and time from admission to operation. The software of SPSS 19.0 was used for statistical analysis.Results:The incidence of lower limb DVT, before operation was 29.92%, including 26 males (24.07%) and 82 females (75.93%). The results of multivariate logistic regression analysis showed that diabetes mellitus ( OR=2.127, 95% CI: 1.134-3.989, P=0.019), coronary heart disease ( OR=1.692, 95% CI: 1.056-2.713, P=0.029) and the time from injury to admission ( OR=1.677, 95% CI: 1.037-2.712, P=0.035) were independent risk factors for DVT in elderly patients undergoing hip arthroplasty. The incidence of lower limb DVT, after operation was 46.54%. After operation, proximal thrombus were occurred in 2 cases (1.19%), distal thrombus in 143 cases (85.12%), and mixed thrombus in 23 cases (13.69%). Postoperative thrombus was ipsilateral to the fracture limb in 84 cases (50.00%), thrombus was located in the healthy side of the fracture in 19 cases (11.31%), and DVT occurred in 65 cases (38.69%) in both lower limbs. Conclusions:Delayed admission longed than 48 hours, coronary heart disease and diabetes mellitus are the risk factors for the formation of DVT. The thrombus that existed before operation and did not disappear after operation accounted for 48.81% of the total incidence of postoperative thrombosis, and the new thrombus accounted for 51.19% of the total incidence of postoperative thrombosis. For the elderly patients with femoral neck fracture undergoing hip arthroplasty, ultrasonic examination of both lower limbs should be performed before and after operation to find the changes of thrombus in time and do a good job of prevention and treatment.

Objective:To establish an artificial intelligence (AI)-assisted diagnostic system for lung cancer via deep transfer learning.Methods:The researchers collected 519 lung pathologic slides from 2016 to 2019, covering various lung tissues, including normal tissues, adenocarcinoma, squamous cell carcinoma and small cell carcinoma, from the Beijing Chest Hospital, the Capital Medical University. The slides were digitized by scanner, and 316 slides were used as training set and 203 as the internal test set. The researchers labeled all the training slides by pathologists and establish a semantic segmentation model based on DeepLab v3 with ResNet-50 to detect lung cancers at the pixel level. To perform transfer learning, the researchers utilized the gastric cancer detection model to initialize the deep neural network parameters. The lung cancer detection convolutional neural network was further trained by fine-tuning of the labeled data. The deep learning model was tested by 203 slides in the internal test set and 1 081 slides obtained from TCIA database, named as the external test set.Results:The model trained with transfer learning showed substantial accuracy advantage against the one trained from scratch for the internal test set [area under curve (AUC) 0.988 vs. 0.971, Kappa 0.852 vs. 0.832]. For the external test set, the transferred model achieved an AUC of 0.968 and Kappa of 0.828, indicating superior generalization ability. By studying the predictions made by the model, the researchers obtained deeper understandings of the deep learning model.Conclusions:The lung cancer histopathological diagnostic system achieves higher accuracy and superior generalization ability. With the development of histopathological AI, the transfer learning can effectively train diagnosis models and shorten the learning period, and improve the model performance.