Background:Distinguishing between primary clear cell carcinoma of the liver(PCCCL)and common hepatocellular carcinoma(CHCC)through traditional inspection methods before the operation is difficult.This study aimed to e...Background:Distinguishing between primary clear cell carcinoma of the liver(PCCCL)and common hepatocellular carcinoma(CHCC)through traditional inspection methods before the operation is difficult.This study aimed to establish a Faster region-based convolutional neural network(RCNN)model for the accurate differential diagnosis of PCCCL and CHCC.Methods:In this study,we collected the data of 62 patients with PCCCL and 1079 patients with CHCC in Beijing YouAn Hospital from June 2012 to May 2020.A total of 109 patients with CHCC and 42 patients with PCCCL were randomly divided into the training validation set and the test set in a ratio of 4:1.The Faster RCNN was used for deep learning of patients’data in the training validation set,and established a convolutional neural network model to distinguish PCCCL and CHCC.The accuracy,average precision,and the recall of the model for diagnosing PCCCL and CHCC were used to evaluate the detection performance of the Faster RCNN algorithm.Results:A total of 4392 images of 121 patients(1032 images of 33 patients with PCCCL and 3360 images of 88 patients with CHCC)were uesd in test set for deep learning and establishing the model,and 1072 images of 30 patients(320 images of nine patients with PCCCL and 752 images of 21 patients with CHCC)were used to test the model.The accuracy of the model for accurately diagnosing PCCCL and CHCC was 0.962(95%confidence interval[CI]:0.931-0.992).The average precision of the model for diagnosing PCCCL was 0.908(95%CI:0.823-0.993)and that for diagnosing CHCC was 0.907(95%CI:0.823-0.993).The recall of the model for diagnosing PCCCL was 0.951(95%CI:0.916-0.985)and that for diagnosing CHCC was 0.960(95%CI:0.854-0.962).The time to make a diagnosis using the model took an average of 4 s for each patient.Conclusion:The Faster RCNN model can accurately distinguish PCCCL and CHCC.This model could be important for clinicians to make appropriate treatment plans for patients with PCCCL or CHCC.展开更多
The Five-hundred-meter Aperture Spherical radio Telescope(FAST) has an active reflector.During observations, the reflector will be deformed into a paraboloid 300 meters in diameter. To improve its surface accuracy, we...The Five-hundred-meter Aperture Spherical radio Telescope(FAST) has an active reflector.During observations, the reflector will be deformed into a paraboloid 300 meters in diameter. To improve its surface accuracy, we propose a scheme for photogrammetry to measure the positions of 2226 nodes on the reflector. The way to detect the nodes in the photos is the key problem in this application of photogrammetry. This paper applies a convolutional neural network(CNN) with candidate regions to detect the nodes in the photos. Experimental results show a high recognition rate of 91.5%, which is much higher than the recognition rate for traditional edge detection.展开更多
相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT...相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT划分延伸而来。新划分方法提高了压缩效率,但导致编码时间急剧增加。为了降低编码复杂度,提出了一种结合深度学习方法和MTT方向早期判决的快速帧内编码算法。首先使用轻量级的卷积神经网络(convolutional neural network,CNN)对QT和部分MTT进行预测划分,其余MTT则采用提前预测MTT划分方向的方法作进一步的优化。实验结果表明,所提方法能够大幅降低编码复杂度,相比于原始编码器的编码时间减少了74.3%,且只有3.3%的码率损失,性能优于对比的方法。展开更多
Learning an effective object detector with little supervision is an essential but challenging problem in computer vision applications. In this paper, we consider the problem of learning a deep convolutional neural net...Learning an effective object detector with little supervision is an essential but challenging problem in computer vision applications. In this paper, we consider the problem of learning a deep convolutional neural network (CNN) based object detector using weakly-supervised and semi-supervised information in the framework of fast region-based CNN (Fast R-CNN). The target is to obtain an object detector as accurate as the fully-supervised Fast R-CNN, but it requires less image annotation effort. To solve this problem, we use weakly-supervised training images (i.e., only the image-level annotation is given) and a few proportions of fully-supervised training images (i.e., the bounding box level annotation is given), that is a weakly-and semi-supervised (WASS) object detection setting. The proposed solution is termed as WASS R-CNN, in which there are two main components. At first, a weakly-supervised R-CNN is firstly trained;after that semi-supervised data are used for finetuning the weakly-supervised detector. We perform object detection experiments on the PASCAL VOC 2007 dataset. The proposed WASS R-CNN achieves more than 85% of a fully-supervised Fast R-CNN's performance (measured using mean average precision) with only 10%of fully-supervised annotations together with weak supervision for all training images. The results show that the proposed learning framework can significantly reduce the labeling efforts for obtaining reliable object detectors.展开更多
文摘Background:Distinguishing between primary clear cell carcinoma of the liver(PCCCL)and common hepatocellular carcinoma(CHCC)through traditional inspection methods before the operation is difficult.This study aimed to establish a Faster region-based convolutional neural network(RCNN)model for the accurate differential diagnosis of PCCCL and CHCC.Methods:In this study,we collected the data of 62 patients with PCCCL and 1079 patients with CHCC in Beijing YouAn Hospital from June 2012 to May 2020.A total of 109 patients with CHCC and 42 patients with PCCCL were randomly divided into the training validation set and the test set in a ratio of 4:1.The Faster RCNN was used for deep learning of patients’data in the training validation set,and established a convolutional neural network model to distinguish PCCCL and CHCC.The accuracy,average precision,and the recall of the model for diagnosing PCCCL and CHCC were used to evaluate the detection performance of the Faster RCNN algorithm.Results:A total of 4392 images of 121 patients(1032 images of 33 patients with PCCCL and 3360 images of 88 patients with CHCC)were uesd in test set for deep learning and establishing the model,and 1072 images of 30 patients(320 images of nine patients with PCCCL and 752 images of 21 patients with CHCC)were used to test the model.The accuracy of the model for accurately diagnosing PCCCL and CHCC was 0.962(95%confidence interval[CI]:0.931-0.992).The average precision of the model for diagnosing PCCCL was 0.908(95%CI:0.823-0.993)and that for diagnosing CHCC was 0.907(95%CI:0.823-0.993).The recall of the model for diagnosing PCCCL was 0.951(95%CI:0.916-0.985)and that for diagnosing CHCC was 0.960(95%CI:0.854-0.962).The time to make a diagnosis using the model took an average of 4 s for each patient.Conclusion:The Faster RCNN model can accurately distinguish PCCCL and CHCC.This model could be important for clinicians to make appropriate treatment plans for patients with PCCCL or CHCC.
基金supported by study on the fusion of total station dynamic tracking measuring and IMU inertial measuring for the feed support measurement in FAST (Grant No. 11503048)the Open Project Program of the Key Laboratory of FAST, NAOC, Chinese Academy of Sciencesthe Key Laboratory of Radio Astronomy, Chinese Academy of Sciences
文摘The Five-hundred-meter Aperture Spherical radio Telescope(FAST) has an active reflector.During observations, the reflector will be deformed into a paraboloid 300 meters in diameter. To improve its surface accuracy, we propose a scheme for photogrammetry to measure the positions of 2226 nodes on the reflector. The way to detect the nodes in the photos is the key problem in this application of photogrammetry. This paper applies a convolutional neural network(CNN) with candidate regions to detect the nodes in the photos. Experimental results show a high recognition rate of 91.5%, which is much higher than the recognition rate for traditional edge detection.
文摘相比于高效视频编码(high efficiency video coding,HEVC)标准,新一代编码标准多功能视频编码(versatile video coding,VVC)引入了很多新的技术,其中包括四叉树(quadtree,QT)和多类型树(multi-type tree,MTT)划分,MTT划分由HEVC中的QT划分延伸而来。新划分方法提高了压缩效率,但导致编码时间急剧增加。为了降低编码复杂度,提出了一种结合深度学习方法和MTT方向早期判决的快速帧内编码算法。首先使用轻量级的卷积神经网络(convolutional neural network,CNN)对QT和部分MTT进行预测划分,其余MTT则采用提前预测MTT划分方向的方法作进一步的优化。实验结果表明,所提方法能够大幅降低编码复杂度,相比于原始编码器的编码时间减少了74.3%,且只有3.3%的码率损失,性能优于对比的方法。
基金This work was supported by the National Natural Science Foundation of China under Grant Nos.61876212,61733007,and 61572207the National Key Research and Development Program of China under Grant No.2018YFB1402604.
文摘Learning an effective object detector with little supervision is an essential but challenging problem in computer vision applications. In this paper, we consider the problem of learning a deep convolutional neural network (CNN) based object detector using weakly-supervised and semi-supervised information in the framework of fast region-based CNN (Fast R-CNN). The target is to obtain an object detector as accurate as the fully-supervised Fast R-CNN, but it requires less image annotation effort. To solve this problem, we use weakly-supervised training images (i.e., only the image-level annotation is given) and a few proportions of fully-supervised training images (i.e., the bounding box level annotation is given), that is a weakly-and semi-supervised (WASS) object detection setting. The proposed solution is termed as WASS R-CNN, in which there are two main components. At first, a weakly-supervised R-CNN is firstly trained;after that semi-supervised data are used for finetuning the weakly-supervised detector. We perform object detection experiments on the PASCAL VOC 2007 dataset. The proposed WASS R-CNN achieves more than 85% of a fully-supervised Fast R-CNN's performance (measured using mean average precision) with only 10%of fully-supervised annotations together with weak supervision for all training images. The results show that the proposed learning framework can significantly reduce the labeling efforts for obtaining reliable object detectors.