Recent advances in 3D fully convolutional networks (FCN) have made it feasible to produce dense voxel-wise predictions of volumetric images. In this work, we show that a multi-class 3D FCN trained on manually labeled CT scans of several anatomical structures (ranging from the large organs to thin vessels) can achieve competitive segmentation results, while avoiding the need for handcrafting features or training class-specific models. To this end, we propose a two-stage, coarse-to-fine approach that will first use a 3D FCN to roughly define a candidate region, which will then be used as input to a second 3D FCN. This reduces the number of voxels the second FCN has to classify to ∼10% and allows it to focus on more detailed segmentation of the organs and vessels. We utilize training and validation sets consisting of 331 clinical CT images and test our models on a completely unseen data collection acquired at a different hospital that includes 150 CT scans, targeting three anatomical organs (liver, spleen, and pancreas). In challenging organs such as the pancreas, our cascaded approach improves the mean Dice score from 68.5 to 82.2%, achieving the highest reported average score on this dataset. We compare with a 2D FCN method on a separate dataset of 240 CT scans with 18 classes and achieve a significantly higher performance in small organs and vessels. Furthermore, we explore fine-tuning our models to different datasets. Our experiments illustrate the promise and robustness of current 3D FCN based semantic segmentation of medical images, achieving state-of-the-art results..

No abstract

The accurate reading of pointer meter is a crucial task in complex environments such as substations, military and aerospace. The current recognition algorithm is mainly used to identify the same type and non-tilt meter, which has limited application in real environment. This paper proposes a novel end-to-end intelligent reading method of pointer meter based on deep learning, which locates the meter and extracts the pointer simultaneously without any prior information. Especially, the pointer is directly and precisely extracted using the designed semi-pointer detection method without any handcrafted features designed in advance, which avoids the accumulated error caused by preprocessing. Based on the extracted panel object, including semi-pointer, panel center and scale characters, the indicated value of the pointer is obtained by a local angle method, which can achieve better performance than the traditional angle method by referring to the neighboring scale lines of the pointer. Experimental results demonstrate that the method is faster and more effective than some common methods. It is worth noting that this study has the advantage of being able to recognize pointer meters in complex environments such as tilt, rotation, blur and illumination. It is acceptable for the actual application requirements in real environment with a recognition accuracy of 99.20% and the average reference error of 0.34%.

周登科 1) , 杨颖 1)* , 朱杰 2) , 王库 2) 1) (中国农业大学信息与电气工程学院 北京 100083) 2) (随锐科技集团股份有限公司 北京 100192)

No abstract