•A long-tail medical image recognition algorithm based on memory features is proposed.•The head and tail categories are processed in a multi-branch manner.•Evaluate the effect of the algorithm on two ...medical imaging datasets and two benchmark long-tailed visual recognition datasets.•Compared with recent studies, better results have been achieved in accuracy, mean class recall, F1-Score, precision, and AUC.
Deep learning algorithms show revolutionary potential in computer-aided diagnosis. These computer-aided diagnosis techniques often rely on large-scale, balanced standard datasets. However, there are many rare diseases in real clinical scenarios, which makes the medical datasets present a highly imbalanced long-tailed distribution, leading to the poor generalization ability of deep learning models. Currently, most algorithms to solve this problem involve more complex modules and loss functions. But for complicated tasks in the medical domain, usually suffer from issues such as increased inference time and unstable performance. Therefore, it is a great challenge to develop a computer-aided diagnosis algorithm for long-tailed medical data.
We proposed the Multi-Branch Network based on Memory Features (MBNM) for Long-Tailed Medical Image Recognition. MBNM includes three branches, where each branch focuses on a different learning task: 1) the regular learning branch learns the generalizable feature representations; 2) the tail learning branch gains extra intra-class diversity for the tail classes through the feature memory module and the improved reverse sampler to improve the classification performance of the tail classes; 3) the fusion balance branch integrates various decision-making advantages and introduces an adaptive loss function to re-balance the classification performance of easy and difficult samples.
We conducted experiments on the multi-disease Ophthalmic OCT datasets with imbalance factors of 98.48 and skin image datasets Skin-7 with imbalance factors of 58.3. The Accuracy, MCR, F1-Score, Precision, and AUC of our model were significantly improved over the strong baselines in the auxiliary diagnosis scenario where the clinical medical data is extremely imbalanced. Furthermore, we demonstrated that MBNM outperforms the state-of-the-art models on the publicly available natural image datasets (CIFAR-10 and CIFAR-100).
The proposed algorithm can solve the problem of imbalanced data distribution with little added cost. In addition, the memory module does not act in the inference phase, thereby saving inference time. And it shows outstanding performance on medical images and natural images with a variety of imbalance factors.
Evidence suggests that the microRNA-181 (miR-181) family performs various roles in the pathophysiology of cerebral ischemia and reperfusion injury (CIRI). MiR-181a has been identified as a critical ...determinant of neuronal survival. Moreover, the significance of miR-181a in controlling neuronal death after CIRI has received little attention. The objective of this study was to assess the role of miR-181a in neuronal cell injury after CIRI. To mimic the in-vitro and in-vivo CIRI, we developed an oxygen-glucose deficiency/reoxygenation (OGD/R) model in SH-SY5Y cells and a transient middle cerebral artery occlusion model in rats. MiR-181a expression was significantly higher in both in-vivo and in-vitro CIRI models. The overexpression of miR-181a increased cell damage and oxidative stress caused by OGD/R, whereas inhibition of miR-181a reduced both. PTEN has also been found to be a direct miR-181a target. PTEN overexpression reduced cell apoptosis and oxidative stress induced by miR-181a upregulation under an OGD/R condition. Furthermore, we found that the rs322931 A allele was related to increased miR-181a levels in IS peripheral blood and higher susceptibility to IS. The current results offer new insights into the understanding of the molecular pathophysiology of CIRI, as well as possible new treatment candidates.
The friction and wear behaviour of several kinds of middle temperature abradable seal coatings used in aircraft turbine engine was investigated. Their abradability was evaluated by sliding worn ...volume. The mechanisms of the sliding wear of the coatings are abrasive wear, adhesive wear and oxidation wear, but the weight of the adhesive wear and abrasive wear is different in different coatings and under different test loads. The results show that the abradability decreases with the increase of the hardness for a given kind of coating. Even if the hardness is close, the abradability is very different in different kinds of coatings. So, only by the hardness can the level of abradability not be not judged and the coating not be chosen and designed. The abradability of M313 type of coating is the best, M310 is close to M601 and M307 with low hardness is fairly good, but M307 with high hardness is the worst.
A high performance liquid chromatographic method for the determination of 28 exogenous medicines and endogenous components in the herbal drink was developed. The samples were extracted ultrasonically ...with methanol-water (70:30, v/v), and the extracts were separated in a Thermo Accucore C18 column (100 mm×4.6 mm, 2.6 μm) with methanol-acetonitrile-20 mmol/L ammonium acetate solution (pH 4.2) as the mobile phases by gradient elution. The flow rate was 1.2 mL/min and the column temperature was 35℃. The detection wavelengths were 254 nm and 220 nm. Quantification analysis was performed by the external standard method. The result showed the compounds had a good linear relationship in the range of 1-100 mg/L, and the correlation coefficients (
) were not less than 0.999. The limits of detection (LODs) of the 28 compounds were 1-10 mg/kg in the liquid sample and 20-200 mg/kg in the solid sample. The average recoveries of the 28 compounds in the liquid and solid samples were in the ranges of 88.8%-118.6% and 92.7%-11
In the repair of large segmental bone defects, bone tissue is often unable to heal due to the destruction of the vascular network near the wound site. An ideal bone repair material should have both ...angiogenic and osteogenic capabilities. To achieve this goal, we used biodegradable mesoporous silica nanoparticles (MSNs) as a delivery vehicle for dexamethasone (DEX), a small-molecule drug that induces osteogenic differentiation. Subsequently, chitosan was covalently modified onto the surface of the nanoparticles by glycidoxypropyltrimethoxysilane (GPTMS) to construct nanoparticulate delivery systems (DEX@chi-MSNs) that induce osteoblast formation. The QK peptide, which mimics the α-helical structure of vascular endothelial growth factor (VEGF) binding to the receptor, was adsorbed to the surface of chitosan-modified MSNs nanoparticles (QK@chi-MSNs) to render them with angiogenic ability. The QK@chi-MSNs can promote the formation of the tubular structure of human umbilical vein endothelial cells (HUVECs) and angiogenesis in vivo, as demonstrated by a chicken embryo chorioallantoic test (CAM) and subcutaneous embedding test. The DEX@chi-MSNs can improve alkaline phosphatase (ALP) activity, mineralized nodule formation, and the expression of osteogenic-related genes and proteins by BMSCs. Furthermore, the ability of bone repair and angiogenesis was evaluated in a critical size skull defect model in rats by using nanocarriers loaded with both DEX and QK (QK/DEX@chi-MSNs). The results of computed tomography (CT) scan, histological examination, and immunofluorescence staining indicated that QK/DEX@chi-MSNs can promote bone formation and angiogenesis in vivo, which has broad application prospects in bone tissue engineering.
The light-weight, high-strength poly-ether-ether-ketone (PEEK) and its composite materials are the ideal candidate for a plethora of engineering applications. However, the inherent high viscosity and ...melting temperature of PEEK pose great challenges for the extrusion-based 3D printing process, leaving a wide knowledge gap on the properties and functionalities of novel PEEK-based composites. In this work, the PEEK/CNT composite filaments with different CNT content ranging from 1 to 7 wt% are prepared by single-screw extruder and subsequently printed with a custom-made high temperature 3D printer. The effects of CNT content and annealing on the mechanical strength and electrical conductivity of the 3D printed parts are investigated in detail. The tensile test result shows that the presence of CNT significantly enhances the mechanical strength of the composite. With 7 wt% CNT content, the maximum tensile strength and elastic modulus reaches 107.7 MPa and 1908.0 MPa, respectively, which amounts to a rise of 27.4% and 17.0% compared with the neat PEEK sample. The annealing treatment is similarly effective in strengthening the printed composites. The addition of CNT is also found to greatly improve the electrical conductivity to the PEEK, reaching 101 S/m for 7 wt% CNT. With such superior electrical conductivity, the printed PEEK/CNT composite sample exhibits outstanding electromagnetic interference (EMI) shielding performance, reaching 32.4 dB. However, additional annealing treatment decreases the electrical conductivity as wells as EMI shielding performance. Such decrease is mainly attributed to the growth of the crystals during annealing, which expels the CNTs to the grain boundaries and leads to remarkable agglomeration, which disconnects the pervasive conductive networks. The finding in this work can provide insights on the new designs of PEEK composite feedstock materials and cultivate new 3D printing strategies targeted specifically for highly-functional, high-strength novel composites.
The wedge-shaped interface transition layer model is proposed for determining the crack growing directions in laminated composites. The Navier equations are given with the material parameters varying ...with the angle in the polar coordinate system. The singularity of the order of −1/2 is proved. An effective solver of the ordinary differential equations is adopted, and the continuous eigenfunctions of the stress, strain and displacement fields near the crack tip are obtained. The crack growing directions can be determined rapidly and accurately based on the obtained results and the relevant fracture criterion. Therefore, a highly effective approach is provided for solving the complicated problem of crack growing directions in the laminated composites.
This paper focuses on the algorithm of denoising for 3D point cloud data in 3D modelling technology. Through the problem that it is difficult to retain the feature points of traditional 3D point ...cloud data denoising and the denoising amplitude is too large, taking the change of Gaussian curvature value of sampling points as the basis of region division, an improved denoising algorithm based on the combination of median denoising algorithm and bilateral denoising algorithm is proposed The algorithm is implemented and simulated by pseudo code. The simulation results show that the 3D point cloud data denoising algorithm based on curvature change not only has a higher ability in denoising effect and feature retention effect, but also has a greater advantage in the execution time of the denoising algorithm.
Reverse engineering is an advanced manufacturing technology, which can design, develop and innovate products. The processing technology of 3D point cloud data is the key of reverse engineering ...technology, and the simplification technology is an important part of data processing, which plays a decisive role in the quality and efficiency of subsequent surface modelling. The advantages and disadvantages of different point cloud data reduction algorithms are analysed. By combining the efficiency of three-dimensional mesh algorithm with the accuracy of curvature reduction method, a second reduction algorithm is proposed. Firstly, the algorithm uses the non-uniform mesh reduction algorithm of octree to simplify it, to ensure the data of feature points is not damaged, and then uses the curvature reduction method of B-spline surface fitting to simplify it. The experiment of tooth point cloud data reduction shows that this method can retain the details of the original data better on the basis of higher reduction ratio.
Recent studies have demonstrated that the long non-coding RNA (lncRNA) GAS5 is closely associated with the onset and progression of several tumor types, including renal cell carcinoma (RCC). This ...study sought to evaluate the relationship between two functional GAS5 polymorphisms (rs145204276 and rs55829688) and the risk for RCC in the Han Chinese population.
The rs145204276 and rs55829688 polymorphisms in the GAS5 promoter region were genotyped in 624 RCC patients and 655 age/sex-matched healthy participants. The association between these polymorphisms and RCC risk was then evaluated using odds ratios (ORs) and corresponding 95% confidence intervals (CIs). Additionally, quantitative RT-PCR was used to determine whether these polymorphisms were associated with changes in the levels of expression of GAS5 in 58 RCC patients.
There were significant differences in the GAS5 rs145204276 polymorphism genotype and allele frequencies between the RCC patients and controls (adjusted OR = 0.73, 95% CI = 0.61- 0.87, P = 1.8×10
). When the study participants were stratified based on age, sex, BMI index, and smoking and drinking history, we found that the rs145204276 del allele was associated with a reduced risk for RCC in nondrinkers (P = 3.3×10
), nonsmokers (P = 3.3×10
), females (P = 3.8×10
), and those who were less than 60 years old (P = 3.3×10
). There was also a significant association between the rs145204276 del allele and elevated expression of GAS5 in RCC patients (P = 0.030).
The results of this study revealed an association between the rs145204276 polymorphism in the GAS5 lncRNA and the risk for the development of RCC, thus representing a potentially viable biomarker for identifying individuals at risk of developing this form of cancer.
PDF
