基于卷积神经网络的单目深度估计

针对利用深度学习方法对街道图像进行深度估计，提出采用语义分割的方法解决深度图出现边界模糊等问题；估计深度通过左右视角图生成视差图进行无监督的训练。在网络模型中添加语义分割层，采取多个空洞卷积并行的结构增加感受野，同时减少了图像下采样的次数，降低了由于下采样带来的信息损失，使得的结果更加准确。这也是在深度估计中首次与空洞卷积相结合增加准确率。通过对KITTI街道数据集进行训练，与现有结果相比，除了增加检测准确性，降低错误率之外，使得效果图中的物体更加清晰，并且在效果图中还保留了一些原模型中被忽视掉的细节信息，将原始图像更加完整的表现出来。     

R2D2: A scalable deep learning toolkit for medical imaging segmentation

Deep learning has gained a significant popularity in recent years thanks to its tremendous success across a wide range of relevant fields of applications, including medical image analysis domain in particular. Although convolutional neural networks (CNNs) based medical applications have been providing powerful solutions and revolutionizing medicine, efficiently training of CNNs models is a tedious and challenging task. It is a computationally intensive process taking long time and rare system resources, which represents a significant hindrance to scientific research progress. In order to address this challenge, we propose in this article, R2D2, a scalable intuitive deep learning toolkit for medical imaging semantic segmentation. To the best of our knowledge, the present work is the first that aims to tackle this issue by offering a novel distributed versions of two well-known and widely used CNN segmentation architectures [ie, fully convolutional network (FCN) and U-Net]. We introduce the design and the core building blocks of R2D2. We further present and analyze its experimental evaluation results on two different concrete medical imaging segmentation use cases. R2D2 achieves up to 17.5× and 10.4× speedup than single-node based training of U-Net and FCN, respectively, with a negligible, though still unexpected segmentation accuracy loss. R2D2 offers not only an empirical evidence and investigates in-depth the latest published works but also it facilitates and significantly reduces the effort required by researchers to quickly prototype and easily discover cutting-edge CNN configurations and architectures.     

基于Spider卷积的三维点云分类与分割网络

针对传统的卷积神经网络（CNN）不能直接处理点云数据，需先将点云数据转换为多视图或者体素化网格，导致过程复杂且点云识别精度低的问题，提出一种新型的点云分类与分割网络Linked-Spider CNN。首先，在Spider CNN基础上通过增加Spider卷积层数以获取点云深层次特征；其次，引入残差网络的思想在每层Spider卷积增加短连接构成残差块；然后，将每层残差块的输出特征进行拼接融合形成点云特征；最后，使用三层全连接层对点云特征进行分类或者利用多层卷积层对点云特征进行分割。在ModelNet40和ShapeNet Parts数据集上将所提网络与PointNet、PointNet++和Spider CNN等网络进行对比实验，实验结果表明，所提网络可以提高点云的分类精度和分割效果，说明该网络具有更快的收敛速度和更强的鲁棒性。     

结合时间戳的指纹密钥数据加解密传输方案

目的 对于生物密钥而言，生物特征数据的安全与生物密钥的管理存储都很关键。为了构造能够应用在通信数据传输场景的生物密钥，同时保证生物特征本身的模糊性与密码学的精确性处于一种相对平衡状态，提出一种基于时间戳与指纹密钥的数据加解密传输方案。方法 利用发送方指纹特征点之间的相对信息，与保密随机矩阵生成发送方指纹密钥；借助通信双方的预先设定数与时间戳，生成接收方恢复指纹密钥时所需的辅助信息；利用发送方指纹密钥加密数据，实现密文数据的传输。结果 本文方法在仿真通信双方数据加解密的实现中，测试再生指纹密钥的识别率（GAR）与误识率（FAR）。通过实验数据分析，表明了本文提出的指纹密钥生成方法的可用性，以及指纹密钥作为数字身份所具备的可认证性，其中真实发送方的再生指纹密钥识别率可高达99.8%，并且本方案还可用于即时通信、对称加密等多种场景当中。结论 本文方法利用时间戳确定了通信事件的唯一性与不可否认性，同时实现了指纹密钥恢复时的"一次一密"。此外，方案通过保密随机矩阵实现了发送方指纹密钥的可撤销，极大程度保障了指纹数据的安全性。     

Multi-path connected network for medical image segmentation

In recent years, deep learning has been successfully applied to medical image segmentation. However, as the network extends deeper, the consecutive downsampling operations will lead to more loss of spatial information. In addition, the limited data and diverse targets increase the difficulty for medical image segmentation. To address these issues, we propose a multi-path connected network (MCNet) for medical segmentation problems. It integrates multiple paths generated by pyramid pooling into the encoding phase to preserve semantic information and spatial details. We utilize multi-scale feature extractor block (MFE block) in the encoder to obtain large and multi-scale receptive fields. We evaluated MCNet on three medical datasets with different image modalities. The experimental results show that our method achieves better performance than the state-of-the-art approaches. Our model has strong feature learning ability and is robust to capture different scale targets. It can achieve satisfactory results while using only 0.98 million (M) parameters.     

miR-181b promotes the oncogenesis of renal cell carcinoma by targeting TIMP3

Renal cell carcinoma (RCC) has a poor prognosis due to limited diagnosis and treatment. Thus, it is necessary to find novel prognostic biomarkers and therapeutic targets. The aberrant expression of microRNAs plays an important role in RCC oncogenesis. Tissue inhibitors of metalloproteinase 3 (TIMP3) acts as a downstream target of miR-181b. The aim of this study was to understand the role and molecular mechanism of miR-181b in RCC oncogenesis. The results showed that miR-181b expression was significantly higher in RCC tumour tissues, especially in those with significant invasion or metastasis. miR-181b overexpression promoted proliferation and migration of the RCC cell line 786-O, while miR-181b knockdown had the opposite effect. In addition, miR-181b was inversely correlated with TIMP3 expression in RCC tumour tissues. miR-181b overexpression reduced TIMP3 expression in RCC cell line 786-O or OS-RC-2, while miR-181b knockdown had the inverse effect. Mechanistically, a luciferase reporter assay confirmed the binding sites of miR-181b on the 3’-UTR of TIMP3, confirming the targeting effect of miR-181b on TIMP3. Overall, miR-181b promotes the development and progression of RCC by targeting TIMP3 expression, indicating the potential use of miR-181b in the diagnosis and treatment of RCC.     

A Feature Selection Strategy to Optimize Retinal Vasculature Segmentation

Diabetic retinopathy (DR) is a complication of diabetes mellitus that appears in the retina. Clinitians use retina images to detect DR pathological signs related to the occlusion of tiny blood vessels. Such occlusion brings a degenerative cycle between the breaking off and the new generation of thinner and weaker blood vessels. This research aims to develop a suitable retinal vasculature segmentation method for improving retinal screening procedures by means of computer-aided diagnosis systems. The blood vessel segmentation methodology relies on an effective feature selection based on Sequential Forward Selection, using the error rate of a decision tree classifier in the evaluation function. Subsequently, the classification process is performed by three alternative approaches: artificial neural networks, decision trees and support vector machines. The proposed methodology is validated on three publicly accessible datasets and a private one provided by Hospital Sant Joan of Reus. In all cases we obtain an average accuracy above 96% with a sensitivity of 72% in the blood vessel segmentation process. Compared with the state-of-the-art, our approach achieves the same performance as other methods that need more computational power. Our method significantly reduces the number of features used in the segmentation process from 20 to 5 dimensions. The implementation of the three classifiers confirmed that the five selected features have a good effectiveness, independently of the classification algorithm.     

Automatic Segmentation and Detection System for Varicocele Using Ultrasound Images

The enlarged veins in the pampiniform venous plexus, known as varicocele disease, are typically identified using ultrasound scans. The medical diagnosis of varicocele is based on examinations made in three positions taken to the right and left testicles of the male patient. The proposed system is designed to determine whether a patient is affected. Varicocele is more frequent on the left side of the scrotum than on the right and physicians commonly depend on the supine position more than other positions. Therefore, the experimental results of this study focused on images taken in the supine position of the left testicles of patients. There are two possible vein structures in each image: a cross-section (circular) and a tube (non-circular) structure. This proposed system identifies dilated (varicocele) veins of these structures in ultrasound images in three stages: preprocessing, processing, and detection and measurement. These three stages are applied in three different color modes: Grayscale, Red-Green-Blue (RGB), and Hue, Saturation, and Value (HSV). In the preprocessing stage, the region of interest enclosing the pampiniform plexus area is extracted using a median filter and threshold segmentation. Then, the processing stage employs different filters to perform image denoising. Finally, edge detection is applied in multiple steps and the detected veins are measured to determine if dilated veins exist. Overall implementation results showed the proposed system is faster and more effective than the previous work.     

Unconstrained Hand Dorsal Veins Image Database and Recognition System

Hand veins can be used effectively in biometric recognition since they are internal organs that, in contrast to fingerprints, are robust under external environment effects such as dirt and paper cuts. Moreover, they form a complex rich shape that is unique, even in identical twins, and allows a high degree of freedom. However, most currently employed hand-based biometric systems rely on hand-touch devices to capture images with the desired quality. Since the start of the COVID-19 pandemic, most hand-based biometric systems have become undesirable due to their possible impact on the spread of the pandemic. Consequently, new contactless hand-based biometric recognition systems and databases are desired to keep up with the rising hygiene awareness. One contribution of this research is the creation of a database for hand dorsal veins images obtained contact-free with a variation in capturing distance and rotation angle. This database consists of 1548 images collected from 86 participants whose ages ranged from 19 to 84 years. For the other research contribution, a novel geometrical feature extraction method has been developed based on the Curvelet Transform. This method is useful for extracting robust rotation invariance features from vein images. The database attributes and the veins recognition results are analyzed to demonstrate their efficacy.