AIM: To determine whether contrast-enhanced ultrasound(CEUS) can improve the precision of breast imaging reporting and data system(BI-RADS) categorization. METHODS: A total of 230 patients with 235 solid breast lesion...AIM: To determine whether contrast-enhanced ultrasound(CEUS) can improve the precision of breast imaging reporting and data system(BI-RADS) categorization. METHODS: A total of 230 patients with 235 solid breast lesions classified as BI-RADS 4 on conventional ultrasound were evaluated. CEUS was performed within one week before core needle biopsy or surgical resection and a revised BI-RADS classification was assigned based on 10 CEUS imaging characteristics. Receiver operating characteristic curve analysis was then conducted to evaluate the diagnostic performance of CEUS-based BI-RADS assignment with pathological examination as reference criteria. RESULTS: The CEUS-based BI-RADS evaluation classified 116/235(49.36%) lesions into category 3, 20(8.51%), 13(5.53%) and 12(5.11%) lesions into categories 4A, 4B and 4C, respectively, and 74(31.49%) into category 5. Selecting CEUS-based BI-RADS category 4A as an appropriate cut-off gave sensitivity and specificity values of 85.4% and 87.8%, respectively, for the diagnosisof malignant disease. The cancer-to-biopsy yield was 73.11% with CEUS-based BI-RADS 4A selected as the biopsy threshold compared with 40.85% otherwise, while the biopsy rate was only 42.13% compared with 100% otherwise. Overall, only 4.68% of invasive cancers were misdiagnosed.CONCLUSION: This pilot study suggests that evaluation of BI-RADS 4 breast lesions with CEUS results in reduced biopsy rates and increased cancer-to-biopsy yields.展开更多
Valuable industrial oil and gas were discovered in the formations of Ordovician, Carboniferous and Triassic of the Tahe （塔河） oilfield, Xinjiang （新疆）, China. The Carboniferous formations contain several oil- an...Valuable industrial oil and gas were discovered in the formations of Ordovician, Carboniferous and Triassic of the Tahe （塔河） oilfield, Xinjiang （新疆）, China. The Carboniferous formations contain several oil- and gas-bearing layers. The lateral distribution of Carboniferous reservoir is unstable, and thin layers are crossbedded. This makes it difficult to do lateral formations＇ contrast and reservoir prediction, so it is necessary to develop a method that can achieve reservoir lateral contrast and prediction by using multi-well logging data and seismic data. To achieve reservoir lateral contrast and prediction at the Carboniferous formations of the Tahe oilfield, processing and interpretation of logging data from a single well were done first. The processing and interpretation include log pretreatment, en- vironmental correction and computation of reservoir＇s parameters （porosity, clay content, water saturation, etc.）. Based on the previous work, the data file of logging information of multi-well was formed, and then the lateral distribution pictures （2D and 3D pictures of log curves and reservoir parameters） can be drawn. Comparing multi-well＇s logging information, seismic profiles and geological information （sedimentary sign）, the reservoir of the Carboniferous in the Tahe oilfield can be contrasted and pre- dicted laterally. The sand formation of Carboniferous can be subdivided. The results of reservoir contrast and prediction of the Carboniferous formations show that 2D and 3D pictures of multi-weU reser- voir parameters make the lateral distribution of reservoir and oil-bearing sand very clear, the connectedness of the reservoir of neighboring wells can be analyzed, and five sand bodies can be identified based on the reservoir＇s lateral distribution, geological information and seismic data.展开更多
AIM: To build and evaluate predictive models for contrast-enhanced ultrasound(CEUS) of the breast to distinguish between benign and malignant lesions. METHODS: A total of 235 breast imaging reporting and data system(B...AIM: To build and evaluate predictive models for contrast-enhanced ultrasound(CEUS) of the breast to distinguish between benign and malignant lesions. METHODS: A total of 235 breast imaging reporting and data system(BI-RADS) 4 solid breast lesions were imaged via CEUS before core needle biopsy or surgical resection. CEUS results were analyzed on 10 enhancing patterns to evaluate diagnostic performance of three benign and three malignant CEUS models, with pathological results used as the gold standard. A logistic regression model was developed basing on the CEUS results, and then evaluated with receiver operating curve(ROC). RESULTS: Except in cases of enhanced homogeneity, the rest of the 9 enhancement appearances were statistically significant(P < 0.05). These 9 enhancement patterns were selected in the final step of the logistic regression analysis, with diagnostic sensitivity and specificity of 84.4% and 82.7%, respectively, and the area under the ROC curve of 0.911. Diagnostic sensitivity, specificity, and accuracy of the malignant vs benign CEUS models were 84.38%, 87.77%, 86.38% and 86.46%, 81.29% and 83.40%, respectively. CONCLUSION: The breast CEUS models can predict risk of malignant breast lesions more accurately, decrease false-positive biopsy, and provide accurate BIRADS classification.展开更多
传统企业合作伙伴推荐方法过度依赖技术特征而忽视多维因素影响。本研究旨在探究企业合作关系的多维影响因素及推荐机制,为企业寻找合适合作伙伴和制定有效创新策略提供技术支持。本研究提出一种基于交叉多头对比学习网络(cross-attenti...传统企业合作伙伴推荐方法过度依赖技术特征而忽视多维因素影响。本研究旨在探究企业合作关系的多维影响因素及推荐机制,为企业寻找合适合作伙伴和制定有效创新策略提供技术支持。本研究提出一种基于交叉多头对比学习网络(cross-attention multi-head contrastive network,CAMC-Net)的企业合作伙伴推荐方法,融合企业、专利和政策数据,通过交叉多头注意力机制建模企业关系的双向互补特性,并引入对比学习策略优化企业表示空间分布。以新能源产业为例,在专利IPC(International Patent Classification)分类号为H02P和H10的企业合作数据集上进行验证,CAMC-Net模型在企业关系识别任务上AUC(area under the curve)分别达到0.9425和0.9251,准确率分别为0.8644和0.8387,F1值分别达到0.8707和0.8471,优于基线模型。通过消融实验证明了政策数据与模型组件的有效性。但现有的研究数据主要基于单一领域,未来需探索跨领域企业合作伙伴推荐方法;同时,模型缺乏对多模态数据的考虑,需要探索更高效的多模态特征融合策略。展开更多
针对异常的存在导致节点邻域信息不可靠的问题,提出一种高效的无监督图异常检测方法。该方法借助邻域增强策略构建多类型的中心节点的邻域集合,捕捉高质量的节点表示,并获取高准确度的自邻相似度。首先,通过优化一个基于动态邻域增强的...针对异常的存在导致节点邻域信息不可靠的问题,提出一种高效的无监督图异常检测方法。该方法借助邻域增强策略构建多类型的中心节点的邻域集合,捕捉高质量的节点表示,并获取高准确度的自邻相似度。首先,通过优化一个基于动态邻域增强的信息提取模块,自适应地选择最优邻域策略,从而克服传统固定邻域选择方法在信息提取过程中特征单一的局限性;其次,为了降低节点特征融合时自身冗余信息的干扰,提出一种匿名消息传递方案,该方案能够隔离节点自身特征,只专注于邻域信息,从而提高消息聚合的质量;最后,通过设计一种自适应的加权异常评分模块,以节点之间距离作为评估尺度来获取节点的异常度,从而细化异常检测结果。在5个数据集上的实验结果表明,所提方法在应对复杂图结构的异常检测方面的表现优于现有主流方法 CoLA(Anomaly detection on attributed networks via Contrastive self-supervised Learning),其中对异常样本的识别能力指标——AUPRC(Area Under the Precision-Recall Curve)至少提升了8.0%。展开更多
文摘AIM: To determine whether contrast-enhanced ultrasound(CEUS) can improve the precision of breast imaging reporting and data system(BI-RADS) categorization. METHODS: A total of 230 patients with 235 solid breast lesions classified as BI-RADS 4 on conventional ultrasound were evaluated. CEUS was performed within one week before core needle biopsy or surgical resection and a revised BI-RADS classification was assigned based on 10 CEUS imaging characteristics. Receiver operating characteristic curve analysis was then conducted to evaluate the diagnostic performance of CEUS-based BI-RADS assignment with pathological examination as reference criteria. RESULTS: The CEUS-based BI-RADS evaluation classified 116/235(49.36%) lesions into category 3, 20(8.51%), 13(5.53%) and 12(5.11%) lesions into categories 4A, 4B and 4C, respectively, and 74(31.49%) into category 5. Selecting CEUS-based BI-RADS category 4A as an appropriate cut-off gave sensitivity and specificity values of 85.4% and 87.8%, respectively, for the diagnosisof malignant disease. The cancer-to-biopsy yield was 73.11% with CEUS-based BI-RADS 4A selected as the biopsy threshold compared with 40.85% otherwise, while the biopsy rate was only 42.13% compared with 100% otherwise. Overall, only 4.68% of invasive cancers were misdiagnosed.CONCLUSION: This pilot study suggests that evaluation of BI-RADS 4 breast lesions with CEUS results in reduced biopsy rates and increased cancer-to-biopsy yields.
基金supported by the Petroleum and Geological Bureau of SINOPEC,China (No. 200002)
文摘Valuable industrial oil and gas were discovered in the formations of Ordovician, Carboniferous and Triassic of the Tahe （塔河） oilfield, Xinjiang （新疆）, China. The Carboniferous formations contain several oil- and gas-bearing layers. The lateral distribution of Carboniferous reservoir is unstable, and thin layers are crossbedded. This makes it difficult to do lateral formations＇ contrast and reservoir prediction, so it is necessary to develop a method that can achieve reservoir lateral contrast and prediction by using multi-well logging data and seismic data. To achieve reservoir lateral contrast and prediction at the Carboniferous formations of the Tahe oilfield, processing and interpretation of logging data from a single well were done first. The processing and interpretation include log pretreatment, en- vironmental correction and computation of reservoir＇s parameters （porosity, clay content, water saturation, etc.）. Based on the previous work, the data file of logging information of multi-well was formed, and then the lateral distribution pictures （2D and 3D pictures of log curves and reservoir parameters） can be drawn. Comparing multi-well＇s logging information, seismic profiles and geological information （sedimentary sign）, the reservoir of the Carboniferous in the Tahe oilfield can be contrasted and pre- dicted laterally. The sand formation of Carboniferous can be subdivided. The results of reservoir contrast and prediction of the Carboniferous formations show that 2D and 3D pictures of multi-weU reser- voir parameters make the lateral distribution of reservoir and oil-bearing sand very clear, the connectedness of the reservoir of neighboring wells can be analyzed, and five sand bodies can be identified based on the reservoir＇s lateral distribution, geological information and seismic data.
文摘AIM: To build and evaluate predictive models for contrast-enhanced ultrasound(CEUS) of the breast to distinguish between benign and malignant lesions. METHODS: A total of 235 breast imaging reporting and data system(BI-RADS) 4 solid breast lesions were imaged via CEUS before core needle biopsy or surgical resection. CEUS results were analyzed on 10 enhancing patterns to evaluate diagnostic performance of three benign and three malignant CEUS models, with pathological results used as the gold standard. A logistic regression model was developed basing on the CEUS results, and then evaluated with receiver operating curve(ROC). RESULTS: Except in cases of enhanced homogeneity, the rest of the 9 enhancement appearances were statistically significant(P < 0.05). These 9 enhancement patterns were selected in the final step of the logistic regression analysis, with diagnostic sensitivity and specificity of 84.4% and 82.7%, respectively, and the area under the ROC curve of 0.911. Diagnostic sensitivity, specificity, and accuracy of the malignant vs benign CEUS models were 84.38%, 87.77%, 86.38% and 86.46%, 81.29% and 83.40%, respectively. CONCLUSION: The breast CEUS models can predict risk of malignant breast lesions more accurately, decrease false-positive biopsy, and provide accurate BIRADS classification.
文摘传统企业合作伙伴推荐方法过度依赖技术特征而忽视多维因素影响。本研究旨在探究企业合作关系的多维影响因素及推荐机制,为企业寻找合适合作伙伴和制定有效创新策略提供技术支持。本研究提出一种基于交叉多头对比学习网络(cross-attention multi-head contrastive network,CAMC-Net)的企业合作伙伴推荐方法,融合企业、专利和政策数据,通过交叉多头注意力机制建模企业关系的双向互补特性,并引入对比学习策略优化企业表示空间分布。以新能源产业为例,在专利IPC(International Patent Classification)分类号为H02P和H10的企业合作数据集上进行验证,CAMC-Net模型在企业关系识别任务上AUC(area under the curve)分别达到0.9425和0.9251,准确率分别为0.8644和0.8387,F1值分别达到0.8707和0.8471,优于基线模型。通过消融实验证明了政策数据与模型组件的有效性。但现有的研究数据主要基于单一领域,未来需探索跨领域企业合作伙伴推荐方法;同时,模型缺乏对多模态数据的考虑,需要探索更高效的多模态特征融合策略。
文摘针对异常的存在导致节点邻域信息不可靠的问题,提出一种高效的无监督图异常检测方法。该方法借助邻域增强策略构建多类型的中心节点的邻域集合,捕捉高质量的节点表示,并获取高准确度的自邻相似度。首先,通过优化一个基于动态邻域增强的信息提取模块,自适应地选择最优邻域策略,从而克服传统固定邻域选择方法在信息提取过程中特征单一的局限性;其次,为了降低节点特征融合时自身冗余信息的干扰,提出一种匿名消息传递方案,该方案能够隔离节点自身特征,只专注于邻域信息,从而提高消息聚合的质量;最后,通过设计一种自适应的加权异常评分模块,以节点之间距离作为评估尺度来获取节点的异常度,从而细化异常检测结果。在5个数据集上的实验结果表明,所提方法在应对复杂图结构的异常检测方面的表现优于现有主流方法 CoLA(Anomaly detection on attributed networks via Contrastive self-supervised Learning),其中对异常样本的识别能力指标——AUPRC(Area Under the Precision-Recall Curve)至少提升了8.0%。
