Multi-phase computed tomography images are widely applied to doctors' routine interpretation of organic lesions or before human abdominal surgery. However, because of the lack of intuitive and three-dimensional (3D) observations, human factors lead to 80% of the clinical errors. A primary malignant tumor resection system based on a virtual reality (VR) helmet and a force-feedback device combined with a 3D printing model is proposed in this paper. First, we used the thin-plate spline (TPS) deformation method to register the different phase images. In the case of a metastatic tumor, a spherical scoring filter (SSF) model was built for searching the tumor pattern with edge detection and subtraction processing, from which the initial tumors were selected on the basis of the calculated score. For hepatocellular carcinoma cases, candidates were extracted as the areas without edges by using edge detection filters on the subtraction image between the equilibrium and arterial phase images. Finally, the false positive (FP) candidate was eliminated before obtaining the 3D shape of the liver tumor in the expansion process, thus providing an accurate volume of lesions for the surgical plan for the liver resection. The results showed that our virtual surgery system enabled the user to simulate the use of a scalpel for cutting and removing the organ labels; 3D display on a VR helmet with a deformable effect; and touching organs with a force-feedback device. Our virtual surgery tools could simulate all of the effects of the doctors' operations and make clinical practice more efficient.

Computer aided diagnosis (CAD) system has been proven to be useful in clinical routine. However, different kinds of software installed in different machine limits the widely usage of CAD by doctors. We transfer our previous CAD system on liver disease into a web based program to enable users to diagnose potential hepatic abnormalities through internet, by using XOJO platform which is easy to make web application under BASIC programming language and provide virtual server function when in running mode. There methods for tumor classification are investigated on web programming: GLCM-ANN, Restricted Boltzmann Machine (RBM) w/o edge computing. The result shows that deep learning has better performance to conventional ANN, but the large weight matrix is a big burden of web response speed without edge computing. Our CAD system can be easily open on different OS and any location with network connection. Such convenience makes diagnosis less time consuming while significantly collecting datasets via internet from different hospitals or even patient him/herself.

Non-small cell lung cancer (NSCLC) is one of the common lung cancers and mainly treated clinically through EGFR-TKI. As time goes on, NSCLC patients are prone to develop resistance to the drug. The resistance may be different types, which increases the difficulties of clinical treatment. In order to capture the biological characteristics of tumor at different timepoints, we analyzed the serial medical images of NSCLC patients, and proposed a neural network using Convolutional Neural Network (CNN) and Recurrent neural networks (RNN), to predict the type of drug resistance in NSCLC patients. The CNN is used to extract the features of tumor images at different timepoints, the RNN will then analyze these features and take the time domain into account. A total of 168 NSCLC patients are split 3:1 into training cohort and test cohort, which are used to train and evaluate the model. In the end, the model achieved 79.16% accuracy on the test set. The results show that the method has obtain preliminary effect in the prediction of drug resistance in NSCLC patients.

目的:设计并实现一套基于虚拟现实的腹部手术仿真系统,探讨其在临床外科应用的可行性、使用者的评价以及相关临床影响.方法:以多时相增强CT影像作为原始数据,采用基于主观阈值和三维标签技术的半自动器官分割方法对腹部器官进行分割,通过3D-Doctor图像处理工具进行三维模型重建,结合虚拟现实技术与力反馈设备构建腹部手术仿真系统,并通过主观和客观的多方面评价指标对系统性能进行综合度量.结果:构建的腹部手术仿真系统在执行虚拟手术时能够保持流畅运行,得到了外科医生的认可,符合临床应用的要求.结论:该系统保证可靠性的同时易于实现,对于临床治疗和手术训练有着积极的意义.

Convolutional neural network based on

Surface defect detection of industrial products has always been an important part of the manufacturing industry. At present,there is a high false detection rate and low efficiency problem of traditional image processing algorithms which easy to be disturbed by complex background. Aiming at the above problems, a method for surface defect detection based on deep learning is proposed. YOLOv3 network adopted in this paper has great advantages in small target recognition and location of target in complex background. In addition, the train-set is effectively extended by elastic deformation and thin-plate spline algorithm. The experiment results show that the scratch recognition rate is as high as 95.8%, the over-judgment rate is 5.4%,and the missed rate is 1.3%.The method can identify the surface defects in a short time, and the average detection time does not exceed 0.4s, which can meet the real-time and precision requirements of industrial applications.