Abstract:The laser speckles based structured light technique performs 3D reconstruction with a single shot, which makes it one of the most important ways to achieve dynamic measurement. However, it is inaccurate, which makes it difficult to meet the needs of accurate measurement. This paper develops a low-cost and miniature 3D vision system for dynamic depth recovery. A high-precision calibration optimization method for binocular camera and an optimal stereo image correction method based on a two-step rotation are proposed, which improves the calibration and measurement accuracy of system’s parameters. Our system consists of a miniature projector and two cameras. A single pattern with pseudo-random speckles is projected onto the surface of the target object and imaged by a stereo camera. Another single pattern with pseudo-random spots is projected onto the surface of the target object. The depth and contour information of the target is recovered by using the feature matching and triangulation principle. After testing, the measurement accuracy of the proposed structured light system is significantly higher than that of the traditional calibration method under the condition of using optimized calibration and rectification parameters. Compared with traditional method, the sphere reconstruction error of the proposed system is 1.05 mm, and the accuracy has been increased by 52.38%. The cylinder reconstruction error is 1.61 mm, and the accuracy has been increased by 63.98%.

Abstract:The traditional metric learning based few-shot image classification methods are task independent, which leads to poor generalization performance of the model on new query tasks. To solve this problem, a taskrelevant image few-shot learning method was proposed in this paper, which can adaptively adjust the feature of support samples according to the query task. Moreover, a variety of regularization methods to address the overfitting problem under severely-limited data scenarios were also investigated. We conduct comprehensive experiments on two popular benchmarks, i.e., miniImageNet and tieredImageNet. The result of 1-shot task on the miniImageNet by the proposed method was 66.05%, and it outperforms the SOTA (state of the art) approaches by 4.29% under the same backbones.

Abstract:Using surface electromyography (sEMG) signals to gesture recognition is a common method. In order to improve the stability and accuracy of gesture recognition, it usually requires to collect more channels of myoelectric signals. However, this would need a high number of electrodes, resulting the increasing of the complexity of myoelectric recognition system. Therefore, using a small number of sEMG electrodes to ensure the performance of gesture recognition has always been an promising direction in the sEMG-based applications. In this study, we designed a portable four-channel sEMG and impedance signal acquisition device that can simultaneously collect sEMG and tissue impedance signal between differential electrode pairs without adding additional sensors and channels. The self-made device was used to collect the hybrid signals of sEMG and tissue impedance for seven classes of hand gesture recognition. The experimental results show that the four-channel fusion information collected by the system could improve the accuracy and stability of gesture recognition. Compared with using EMG only, the fusion method could improve gesture recognition performance by more than 3% and achieve a recognition rate of 96.2%.

Abstract:Lentiviral vector is an important gene transduction tool for gene therapy and plays an a crucial role in the treatment of hematologic tumors by chimeric antigen receptor (CAR) T cells. The prerequisite for the preparation of high quality CAR-T cells is to obtain lentivirus with high titer by concentration, and ultracentrifugation and ultrafiltration are two main methods for virus enrichment. In order to determine the efficiency of ultracentrifugation and ultrafiltration for virus concentration, and their effect on virus activity in the preparation of research-grade CAR-T, two lentivirus plasmids (GFP and CD19 CAR(CAR19)-mCherry) were selected to package lentivirus, then titer and recovery rate were compared. The results showed that the titer of virus produced by ultrafiltration reached 1011 TU/mL, which was significantly higher than that by ultracentrifugation (P＜0.05), and the higher recovery efficiency was also obtained by ultrafiltration. These results can provide direct theoretical reference for the lentivirus concentration method to support the preparation of laboratory-grade CAR-T, and are helpful for the rapid development of CAR-T basic research.

Abstract:Rapid development of ultrafast ultrasound imaging method has led to novel ultrasound imaging modalities. These novel imaging modalities significantly improve the accuracy and specificity on disease diagnosis and they all need high imaging fame rate to enable accurate tracking of transient tissue variation. However, current ultrafast imaging methods are all based on linear array, convex array or planar array, they haven’t been fully used on endoscopic ultrasonic (EUS) circular array. Therefore, it is still difficult to implant these methods on EUS. To solve this problem, three kinds of ultrafast imaging methods with EUS circular array were studied. The theoretical analysis on image reconstruction and simulation experiments were implemented. The results show that all the methods can improve the imaging frame to nearly 1 kHz with lateral resolution not wider than 2 mm.

Abstract:The human intestinal tract has a complex physiological microenvironment, such as intestinal epithelial cells forming villi structure, fluid shear stress, intestinal peristalsis and other mechanical conditions. Moreover, a large number of microorganisms colonized in the intestinal tract which are closely related to human health. Traditional in vitro cell culture methods can not simulate the complex physiological microenvironment of the intestinal tract. In this research, we designed a microfluidic chip by culturing intestinal cells on the villous basement membrane and combining with fluid shear force to simulate the structure and function of human intestinal tract. The results showed that the intestine chip reproduced the intestinal villi structure and barrier function, increased the expression of mucus, and obtained real-time observation of intestinal bacteria on the intestine chip, which can be used as a powerful research tool for studying the interaction between intestinal microorganisms and the host in vitro.

Abstract:For the problem that doctors manually segment the liver tumor from CT image is time-consuming, labor-intensive, and susceptible to subjective judgment, we propose a deeply supervised residual Unet (DSResUnet) that incorporates residual link and deep supervision into Unet for more precise segmentation. The proposed method was evaluated on the public MICCAI 2017 liver segmentation (LiTS) challenge dataset with Dice coefficient, Jaccard coefficient, average symmetrical surface distance (ASSD), 95% Hausdorff distance (HD95), precision and recall. The experimental results show that the results on the above 7 evaluation indicators of liver segmentation with the proposed DS-ResUnet are 96.06%, 95.08%, 92.54%, 1.98 mm, 12.87 mm, 96.11%, and 96.06%, respectively, achieve superior results on almost all metrics to the widely-used Unet (95.71%, 94.52%, 91.91%, 2.41 mm, 14.21 mm, 95.48%, 96.01%). The results on the above 7 evaluation indicators of liver tumor segmentation with the proposed DS-ResUnet are 67.51%, 76.65%, 54.21%, 6.65 mm, 25.34 mm, 80.39%, and 64.27%, respectively, also better than that of the Unet (60.67%, 73.47%, 47.39%, 9.43 mm, 39.38 mm, 79.61%, 58.01%). Therefore, the proposed DS-ResUnet improves the segmentation results and achieves automatic segmentation of liver and liver tumor regions from the 3D abdominal enhanced CT image.