In Silico Study on Tumor-Size-Dependent Thermal Profiles inside an Anthropomorphic Female Breast Phantom Subjected to Multi-Dipole Antenna Array

Electromagnetic hyperthermia as a potent adjuvant for conventional cancer therapies can be considered valuable in modern oncology, as its task is to thermally destroy cancer cells exposed to high-frequency electromagnetic fields. Hyperthermia treatment planning based on computer in silico simulations has the potential to improve the localized heating of breast tissues through the use of the phased-array dipole applicators. Herein, we intended to improve our understanding of temperature estimation in an anatomically accurate female breast phantom embedded with a tumor, particularly when it is exposed to an eight-element dipole antenna matrix surrounding the breast tissues. The Maxwell equations coupled with the modified Pennes’ bioheat equation was solved in the modelled breast tissues using the finite-difference time-domain (FDTD) engine. The microwave (MW) applicators around the object were modelled with shortened half-wavelength dipole antennas operating at the same 1 GHz frequency, but with different input power and phases for the dipole sources. The total input power of an eight-dipole antenna matrix was set at 8 W so that the temperature in the breast tumor did not exceed 42 °C. Finding the optimal setting for each dipole antenna from the matrix was our primary objective. Such a procedure should form the basis of any successful hyperthermia treatment planning. We applied the algorithm of multi for multi-objective optimization for the power and phases for the dipole sources in terms of maximizing the specific absorption rate (SAR) parameter inside the breast tumor while minimizing this parameter in the healthy tissues. Electro-thermal simulations were performed for tumors of different radii to confirm the reliable operation of the given optimization procedure. In the next step, thermal profiles for tumors of various sizes were calculated for the optimal parameters of dipole sources. The computed results showed that larger tumors heated better than smaller tumors; however, the procedure worked well regardless of the tumor size. This verifies the effectiveness of the applied optimization method, regardless of the various stages of breast tumor development.     

Performance of near-infrared dyes as effective contrast agents for breast cancer detection through simulation of photoacoustic imaging

ABSTRACT

Targeted photoacoustic imaging using exogenous contrast agents can potentially improve early detection of breast cancer, even at significant depths inside the breast. In this study, computer simulations were performed to compare the photoacoustic performance of 11 different near-infrared (NIR) dyes for detecting tumours deep inside the breast tissue. It was observed that the three high performing NIR dyes produced at least two-fold contrast enhancement of a spherical breast tumour embedded at 4?cm depth inside the breast than those of the corresponding endogenous contrast agents. These three selected dyes were employed to visualize small blood vessels deep inside the breast tissue. Although methylene blue provided the best contrast in visualizing tumour blood vessels at depths beyond 3?cm, considering other factors such as availability of suitable targeting agent, indocyanine green at 800?nm may be preferred over all other dyes for deep breast imaging applications.     

On Demand Solid Texture Synthesis Using Deep 3D Networks

This paper describes a novel approach for on demand volumetric texture synthesis based on a deep learning framework that allows for the generation of high-quality three-dimensional (3D) data at interactive rates. Based on a few example images of textures, a generative network is trained to synthesize coherent portions of solid textures of arbitrary sizes that reproduce the visual characteristics of the examples along some directions. To cope with memory limitations and computation complexity that are inherent to both high resolution and 3D processing on the GPU, only 2D textures referred to as ‘slices’ are generated during the training stage. These synthetic textures are compared to exemplar images via a perceptual loss function based on a pre-trained deep network. The proposed network is very light (less than 100k parameters), therefore it only requires sustainable training (i.e. few hours) and is capable of very fast generation (around a second for 256³ voxels) on a single GPU. Integrated with a spatially seeded pseudo-random number generator (PRNG) the proposed generator network directly returns a color value given a set of 3D coordinates. The synthesized volumes have good visual results that are at least equivalent to the state-of-the-art patch-based approaches. They are naturally seamlessly tileable and can be fully generated in parallel.     

基于深度学习的安全缺陷报告预测方法实证研究

软件安全问题的发生在大多数情况下会造成非常严重的后果，及早发现安全问题，是预防安全事故的关键手段之一.安全缺陷报告预测可以辅助开发人员及早发现被测软件中潜藏的安全缺陷，从而尽早得以修复.然而，由于安全缺陷在实际项目中的数量较少，而且特征复杂（即安全缺陷类型繁多，不同类型安全缺陷特征差异性较大），这使得手工提取特征相对困难，并随后造成传统机器学习分类算法在安全缺陷报告预测性能方面存在一定的瓶颈.针对该问题，提出基于深度学习的安全缺陷报告预测方法，采用深度文本挖掘模型TextCNN和TextRNN构建安全缺陷报告预测模型；针对安全缺陷报告文本特征，使用skip-grams方式构建词嵌入矩阵，并借助注意力机制对TextRNN模型进行优化.所构建的模型在5个不同规模的安全缺陷报告数据集上展开了大规模实证研究，实证结果表明：深度学习模型在80%的实验案例中都要优于传统机器学习分类算法，性能指标F1-score平均可提升0.258，在最好的情况下甚至可以提升0.535.除此之外，针对安全缺陷报告数据集存在的类不均衡问题，对不同采样方法进行了实证研究，并对结果进行了分析.     

对抗样本生成技术综述

如今,深度学习已被广泛应用于图像分类和图像识别的问题中,取得了令人满意的实际效果,成为许多人工智能应用的关键所在.在对于模型准确率的不断探究中,研究人员在近期提出了“对抗样本”这一概念.通过在原有样本中添加微小扰动的方法,成功地大幅度降低原有分类深度模型的准确率,实现了对于深度学习的对抗目的,同时也给深度学习的攻方提供了新的思路,对如何开展防御提出了新的要求.在介绍对抗样本生成技术的起源和原理的基础上,对近年来有关对抗样本的研究和文献进行了总结,按照各自的算法原理将经典的生成算法分成两大类——全像素添加扰动和部分像素添加扰动.之后,以目标定向和目标非定向、黑盒测试和白盒测试、肉眼可见和肉眼不可见的二级分类标准进行二次分类.同时,使用MNIST数据集对各类代表性的方法进行了实验验证,以探究各种方法的优缺点.最后总结了生成对抗样本所面临的挑战及其可以发展的方向,并就该技术的发展前景进行了探讨.     

基于阶段注意力机制的电力线提取算法

电力系统维护是电力系统稳定运行的重要保障，应用智能算法的无人机电力巡检则为电力系统维护提供便捷。电力线提取是自主电力巡检以及保障飞行器低空飞行安全的关键技术，结合深度学习理论进行电力线提取是电力巡检的重要突破点。本文将深度学习方法用于电力线提取任务，结合电力线图像特点嵌入改进的图像输入策略和注意力模块，提出一种基于阶段注意力机制的电力线提取模型（SA-Unet）。本文提出的SA-Unet模型编码阶段采用阶段输入融合策略（Stage input fusion strategy， SIFS），充分利用图像的多尺度信息减少空间位置信息丢失。解码阶段通过嵌入阶段注意力模块（Stage attention module，SAM）聚焦电力线特征，从大量信息中快速筛选出高价值信息。实验结果表明，该方法在复杂背景的多场景中具有良好的性能。     

No-reference quality assessment of dynamic sports videos based on a spatiotemporal motion model

This paper proposes an approach to improve the performance of no-reference video quality assessment for sports videos with dynamic motion scenes using an efficient spatiotemporal model. In the proposed method, we divide the video sequences into video blocks and apply a 3D shearlet transform that can efficiently extract primary spatiotemporal features to capture dynamic natural motion scene statistics from the incoming video blocks. The concatenation of a deep residual bidirectional gated recurrent neural network and logistic regression is used to learn the spatiotemporal correlation more robustly and predict the perceptual quality score. In addition, conditional video block-wise constraints are incorporated into the objective function to improve quality estimation performance for the entire video. The experimental results show that the proposed method extracts spatiotemporal motion information more effectively and predicts the video quality with higher accuracy than the conventional no-reference video quality assessment methods.     

A recurrent neural network using historical data to predict time series indoor PM2.5 concentrations for residential buildings

Due to the severe outdoor PM2.5 pollution in China, many people have installed air-cleaning systems in homes. To make the systems run automatically and intelligently, we developed a recurrent neural network (RNN) that uses historical data to predict the future indoor PM2.5 concentration. The RNN architecture includes an autoencoder and a recurrent part. We used data measured in an apartment over the course of an entire year to train and test the RNN. The data include indoor/outdoor PM2.5 concentration, environmental parameters and time of day. By comparing three different input strategies, we found that a strategy employing historical PM2.5 and time of day as inputs performed best. With this strategy, the model can be applied to predict the relatively stable trend of indoor PM2.5 concentration in advance. When the input length is 2 h and the prediction horizon is 30 min, the median prediction error is 8.3 µg/m³ for the whole test set. For times with indoor PM2.5 concentrations between (20,50] µg/m³ and (50,100] µg/m³, the median prediction error is 8.3 and 9.2 µg/m³, respectively. The low prediction error between the ground-truth and predicted values shows that the RNN can predict indoor PM2.5 concentrations with satisfactory performance.