Dual-Path Vision Transformer用于急性缺血性脑卒中辅助诊断

急性缺血性脑卒中是由于脑组织血液供应障碍导致的脑功能障碍,数字减影脑血管造影(DSA)是诊断脑血管疾病的金标准。基于患者的正面和侧面DSA图像,对急性缺血性脑卒中的治疗效果进行分级评估,构建基于Vision Transformer的双路径图像分类智能模型DPVF。为了提高辅助诊断速度,基于EdgeViT的轻量化设计思想进行了模型的构建;为了使模型保持轻量化的同时具有较高的精度,提出空间-通道自注意力模块,促进Transformer模型捕获更全面的特征信息,提高模型的表达能力;此外,对于DPVF的两分支的特征融合,构建交叉注意力模块对两分支输出进行交叉融合,促使模型提取更丰富的特征,从而提高模型表现。实验结果显示DPVF在测试集上的准确率达98.5%,满足实际需求。

基于双通道Residual-LSTM的SINS/GNSS组合导航算法

针对全球导航卫星系统信号中断情况下SINS/GNSS组合导航系统无法持续进行误差校正的问题,提出一种基于双通道Residual-LSTM的SINS/GNSS组合导航算法。首先,考虑到SINS经度、纬度误差传播特性不同所导致的模型输入、输出信息之间的非线性相关性差异化,构建具有不同权重系数的双通道长短期记忆神经网络模型结构,并引入遗忘信息共享机制自适应地利用历史导航数据对经度、纬度信息进行拟合预测。其次,针对深层神经网络存在的模型退化和梯度消失问题,在多层双通道LSTM网络之间建立残差高速通道形成Residual-LSTM模型结构,以增加不同网络层次之间的信息传播路径。最后,通过实船数据验证本文所提算法的有效性。实验结果表明,与基于常规智能方法的SINS/GNSS组合导航算法相比,所提组合导航算法在GNSS信号中断期间经度误差降低了51.97%,纬度误差降低了31.45%。

U-Net Based Dual-Pooling Segmentation of Bone Metastases in Thoracic SPECT Bone Scintigrams

In order to enhance the performance of the CNN-based segmentation models for bone metastases, this study proposes a segmentation method that integrates dual-pooling, DAC, and RMP modules. The network consists of distinct feature encoding and decoding stages, with dual-pooling modules employed in encoding stages to maintain the background information needed for bone scintigrams diagnosis. Both the DAC and RMP modules are utilized in the bottleneck layer to address the multi-scale problem of metastatic lesions. Experimental evaluations on 306 clinical SPECT data have demonstrated that the proposed method showcases a substantial improvement in both DSC and Recall scores by 3.28% and 6.55% compared the baseline. Exhaustive case studies illustrate the superiority of the methodology.

磁共振酰胺质子转移模态的胶质瘤IDH基因分型识别:基于深度学习的Dual-Aware框架

目的基于磁共振酰胺质子转移(APT)模态数据,提出Dual-Aware深度学习框架实现胶质瘤异柠檬酸脱氢酶(IDH)基因分型,实现无创性的辅助诊断。方法共收集118例多模态磁共振脑影像数据,其中68例为野生型,50例为突变型,所有数据完成了脑部胶质瘤ROI区域勾画。APT模态成像无需造影剂,它对肿瘤的信号强度跟肿瘤恶性程度呈正相关,且对IDH野生型的信号强度高于对IDH突变型的信号强度。针对APT模态肿瘤成像及衍生区域形态多变,且相对其他模态更缺乏明显的边缘轮廓特点,构建Dual-Aware框架,引入Multi-scaleAware模块挖掘多尺度特征,EdgeAware模块挖掘边缘特征,以实现基因分型的自动识别。结果两类Aware机制的引入能够有效提升模型对胶质瘤IDH基因分型的识别率。对各模态数据在准确率和AUC方面都取得了一定的提升,并在APT模态上取得了最佳性能,预测精度达到83.1%,AUC达到0.822,从而验证了APT模态对于脑胶质瘤IDH基因分型识别的优势和有效性。结论本文采用的深度学习算法模型基于APT模态的图像特点构建,验证了APT模态在胶质瘤IDH基因分型识别任务中的有效性,有望在该任务利用APT模态取代目前常用的T1CE模态,避免造影剂注射,以实现无创性的IDH基因分型。

Electron Shape Calculated for the Dual-Charge Dual-Mass Model

A model for the internal structure of the electron using classical physics equations has been previously published by the author. The model employs both positive and negative charges and positive and negative masses. The internal attributes of the electron structure were calculated for both ring and spherical shapes. Further examination of the model reveals an instability for the ring shape. The spherical shape appears to be stable, but relies on tensile or compressive forces of the electron material for stability. The model is modified in this document to eliminate the dependency on material forces. Uniform stability is provided solely by balancing electrical and centrifugal forces. This stability is achieved by slightly elongating the sphere along the spin axis to create a prolate ellipsoid. The semi-major axis of the ellipsoid is the spin axis of the electron, and is calculated to be 1.20% longer than the semi-minor axis, which is the radius of the equator. Although the shape deviates slightly from a perfect sphere, the electric dipole moment is zero. In the author’s previously published document, the attributes of the internal components of the electron, such as charge and mass, were calculated and expressed as ratios to the classically measured values for the composite electron. It is interesting to note that all of these ratios are nearly the same as the inverse of the Fine Structure Constant, with differences of less than 15%. The electron model assumed that the outer surface charge was fixed and uniform. By allowing the charge to be mobile and the shape to have a particular ellipticity, it is shown that the calculated charge and mass ratios for the model can be exactly equal to the Fine Structure Constant and the Constant plus one. The electron radius predicted by the model is 15% greater than the Classical Electron Radius.

Dual-Delay-Path Ring Oscillator with Self-Biased Delay Cells for Clock Generation

This work summarizes the structure and operating features of a high-performance 3-stage dual-delay-path (DDP) voltage-controlled ring oscillator (VCRO) with self-biased delay cells for Phase-Locked Loop (PLL) structurebased clock generation and digital system driving. For a voltage supply VDD = 1.8 V, the resulting set of performance parameters include power consumption PDC = 4.68 mW and phase noise PN@1MHz = -107.8 dBc/Hz. From the trade-off involving PDC and PN, a system level high performance is obtained considering a reference figure-of-merit ( FoM = -224 dBc/Hz ). Implemented at schematic level by applying CMOS-based technology (UMC L180), the proposed VCRO was designed at Cadence environment and optimized at MunEDA WiCkeD tool.

复杂水下环境中的Dual-HFM速度谱优化

在理想环境下,双双曲调频(dual hyperbolic frequency modulation,Dual-HFM)速度谱估计方法可得到高分辨率的连续多普勒估计结果。然而由多径、双目标环境引起的旁瓣干扰,削弱了速度谱方法抑制噪声的能力,影响目标参数估计。针对该方法在多径、双目标等各种复杂水声环境中的应用,进行了深入的讨论,推导了多径、双目标造成的速度谱旁瓣位置,并提出了基于多帧信号的速度谱旁瓣抑制方法,利用另一维度信号空间中目标回波信息与多径杂波旁瓣的差异性,抑制了复杂水下环境中的速度谱旁瓣,并保留了速度谱计算量低的优点。通过数值仿真验证了所提方法的适用性,为低信噪比、多径、双目标环境下的多普勒估计提供了理论依据。

融合时间序列趋势的Dual-ESN机组负荷预测模型

针对传统模型在机组负荷预测中无法充分捕获内部多变量演化模式的问题,提出了一种基于时间序列的趋势和数值信息融合的双重回声状态网络Dual-ESN(dual-echo state network)机组负荷动态预测模型。首先,引入最小二乘法,对相关的多元历史信息按照局部时间跨度进行趋势拟合。进一步,得到有关过程变化的模式序列,并和原本的数值分别被送入两个独立的储备池,以并行的时间维度进行特征学习。其次,将隐层的高维空间状态送入输出层,融合信息,得到所需要的预测结果。最后,基于山西某工厂660 MW机组装置的真实数据集,进行验证。对比已有预测方法,结果表明所提预测模型在多种性能指标上均有提升。

A Novel CMOSDual-Modulus Prescaler Based on New Optimized Structure and Dynamic Circuit Technique

s:A divide- by- 12 8/ 12 9or6 4/ 6 5 dual- modulus prescaler based on new optimized structure and dynam ic circuit technique im plem ented in 0 .2 5 μm CMOS digital technology is described.New optimized structure reduces the propagation delay and has higher operating speed.Based on this structure,an im proved D- flip- flop(DFF) using dynam ic circuit technique is proposed.A prototype is fabricated and the measured results show that this prescaler works well in gigahertz frequency range and consumes only35 m W(including three power- hungry output buffers) when the input frequency is2 .5 GHz and the power supply voltage is2 .5 V.Due to its excellent perform ance,the prescaler could be applied to many RF system s.

Response simulation and theoretical calibration of a dual- induction resistivity LWD tool

In this paper, responses of a new dual-induction resistivity logging-while-drilling （LWD） tool in 3D inhomogeneous formation models are simulated by the vectorfinite element method （VFEM）, the influences of the borehole, invaded zone, surroundingstrata, and tool eccentricity are analyzed, and calibration loop parameters and calibrationcoefficients of the LWD tool are discussed. The results show that the tool has a greater depthof investigation than that of the existing electromagnetic propagation LWD tools and is moresensitive to azimuthal conductivity. Both deep and medium induction responses have linearrelationships with the formation conductivity, considering optimal calibration loop parametersand calibration coefficients. Due to the different depths of investigation and resolution, deepinduction and medium induction are affected differently by the formation model parameters,thereby having different correction factors. The simulation results can provide theoreticalreferences for the research and interpretation of the dual-induction resistivity LWD tools.

基于FPGA内部Dual-RAM的多轴机器人指令存储

为了解决高精度多轴机器人控制系统中FPGA块RAM容量有限的问题,提出了一种保存规划运动指令的方法。利用FPGA块RAM生成Dual-RAM,通过轴选判断器选择出有效运动参数,再按轴号分别存入各Dual-RAM。实验结果表明:采用该方法,直接使用FPGA内部RAM,避免了外部RAM扩展,从而精简了指令存储电路,增强了系统可移植性。该机器人系统目前已成功应用于工件上下料作业,系统运行稳定,点重复精度达0.02 mm,水平联动速度达到5.2 m/s。

Superiority of Digital Subtraction for Analysis of Simultaneously-Acquired Dual-Radiopharmaceutical Parathyroid Scintigraphy

With dual-radiopharmaceutical (DR) parathyroid scintigraphy, imaging with 99mTcO4-or 123I-NaI is combined with 99mTc-sestamibi scintigraphy for localization of parathyroid adenomas. The set images are then either visually compared or digitally subtracted to aid in interpretation. While both EANM and SNMMI guidelines recommend use of digital subtraction over visual interpretation alone, to date, the few formal comparisons performed have not demonstrated superiority. The purpose of this investigation is to rigorously assess the added value of digital subtraction over visual interpretation alone using simultaneously-acquired 123I-NaI and 99mTc-sestamibi images. Materials: 90 consecutive patients with DR parathyroid scintigraphy for primary hyperparathyroidism who underwent successful parathyroidectomy were included. DR planar acquisition was performed 15 minutes post injection using 10% dual energy windows. Digital subtraction was subsequently performed using commercially available software. Images were independently reviewed by 3 nuclear medicine trainees and 2 experienced nuclear medicine physicians with and without digital subtraction. Results were compared with surgical and histopathologic findings, which served as ground truth. Results: 90 patients had a total of 91 confirmed parathyroid lesions. All 5 readers had significantly greater sensitivity with digital subtraction compared with visual interpretation alone while specificity was not significantly diminished. Area under the ROC curve was significantly greater with digital subtraction in 3 of 5 readers. Agreement was greater among trainees and experienced physicians when using digital subtraction. Conclusion: Using an optimized DR planar co-imaging technique, digital subtraction significantly improved inter-observer agreement and confidence of interpretation and increased sensitivity, without diminishing specificity.

一种新型无桥Dual-Boost PFC软开关控制技术

研究了一种非最小电压应力无源无损PWM变换器,与具有最小电压应力的PWM变换器相比,它具有电流应力更小、效率更高的优势。分析了非最小电压应力的无源无损电路网络在无桥Dual-Boost功率因数矫正(Power Factor Correction,PFC)的工作模态,并且给出了相应的参数设计。仿真结果证实了方案的可行性。

基于Dual-Chord模型的资源定位技术研究

为了更高效地实现资源定位,提出了一种基于Dual-Chord的资源定位模型,在该模型中将网络分成主网和子网两层,主网采用k-高频词搜索路由算法,子网采用双向查询Chord算法.理论上得出使用该模型的用户可以实现高效的查全率、查准率,实现高度的资源本地化等.

Application of an Advanced Image-Based Virtual Monoenergetic Reconstruction of Dual Source Dual-Energy CT Data at Low keV Increases Image Quality for Esophageal Cancer Patients Imaging

Purpose: To explore the significance of dual-source computed tomography (DECT) virtual monoenergetic reconstructions technology in improving the imaging quality of esophageal cancer patients. Materials and methods: 68 patients with clinically suspected esophageal cancer (all confirmed by pathology) were collected. Routine plain scan was performed with SIEMENS Force dual-energy CT and then dual-phase scans were performed. The venous phase images were respectively subjected to traditional virtual monoenergetic reconstructions (Mono_E) and new generation virtual monoenergetic reconstructions (Mono+). Mono_E 55 keV and Mono+ 55 keV virtual single-energy images were obtained respectively. The signal-to-noise ratio (SNR) of normal esophageal tissue and esophageal cancer lesions, noise and contrast noise ratio (CNR tumor) of normal esophageal tissue and esophageal cancer lesions were compared among 100 kV images, Mono_E images and Mono+ images. At the same time, two imaging physicians read the films and scored the images of each group by using a 5-point scoring method. Results: Mono+ 55 keV images, SNR, SNRtumor, noise and CNRtumor were statistically different from those of 100 kV images and Mono_E images (P < 0.05). And Mono+ 55 keV images also had the highest subjective score, with statistical significance (P 55 keV images had the best quality. Conclusion: The new generation of virtual monoenergetic reconstructions post-processing (Mono+) could reduce image noise and improve the contrast between esophageal cancer lesions and normal esophageal tissues, which was of great significance to improve the imaging quality of esophageal cancer patients and improve the early detection rate of esophageal cancer.

一种Dual-LSTM混合模型的产线设备状态预测方法与应用

智能化生产线设备健康状态、加工过程状态、产品信息等数据具有复杂化、多样化、大容量的特点,传统上对于设备的运行状况主要依靠人工经验来判断,不能及时有效地给出维护意见.针对上述问题,本文提出了一种基于Dual-LSTM(Long Short-Term M emory)混合模型的时序数据预测方法.首先建立LSTM预测模型对设备状态进行初步预测,然后针对多步预测过程中出现的误差"累积"问题,通过预测残差数据建立LSTM辅助模型对初步预测结果进行修正,最后采用循环迭代的方式实现了对数据的多步预测过程.通过与单LSTM模型进行实验对比,该方法在数据的单步预测和多步预测中的表现均优于单模型,验证了所提方法在时序数据预测上的准确性,为分析生产线整体的运行状态趋势提供了有效地判断依据.

内镜下Dual-knife开窗术在困难胆管插管中的应用

目的探讨Dual-knife在内镜下逆行胰胆管造影术(endoscopic retrograde cholangiopan-creatography,ERCP)胆总管困难插管中应用的安全性和有效性。方法回顾性分析华中科技大学同济医学院附属协和医院2014年3月至2019年8月间2564例行ERCP病人的临床资料,共215例病人在术中被诊断为胆总管插管困难,应用Dual-knife行乳头开窗术(Dual-knife fistulosphincterotomy,DKF)。结果所有215例病人中,212例实现胆管插管,成功率为98.60%;平均时间为3.5 min(2~8 min)。与标准插管相比,并发症发生率无明显增加(P>0.05)。15例病人发生轻度切缘出血,经局部喷洒去甲肾上腺素、Dual-knife凝血、和谐夹夹闭后成功止血;3例病人因开窗后出现潜在穿孔,但未造成后腹膜腔感染,遂保守治疗好转。Dual-knife开窗术后,共有21例(9.8%)出现轻、中度胰腺炎,15例病人出现胆管炎(6.9%),均保守治疗成功;没有病人因穿孔行手术治疗,也没有与该手术相关的死亡。结论采用Dual-knife行乳头开窗术,可实现标准插管失败病人选择性胆管插管;其并不增加病人术后并发症和死亡率,是一种有效、简便、安全的方法。

基于Dual-Tree CWT和自适应双边滤波器的图像去噪算法

针对目前图像去噪方法主要局限于单一噪声,无法有效解决多种混合噪声的不足,提出了一种基于DualTree CWT和自适应双边滤波器的图像去噪算法。该算法使用双树复小波变换对含噪图像进行多尺度和多方向的分解,由改进阈值对各个方向子带的高频系数进行阈值量化,同时由自适应双边滤波对每尺度下低频子带系数进行滤波,并将重构得到的图像进一步去除噪声。实验仿真结果表明,该方法对混合噪声的滤除效果明显优于现有算法,且能较好地保护图像的边缘细节信息,通过客观评价指标峰值信噪比(PSNR)和均方根误差(RMSE)定量比较,PSNR提升了5.333 2~6.527 8 d B,RMSE可降低29.41%~46.03%,运行时间仅为1.492 0 s,整体降噪性能更优。

CMOS Dual-Band Low Noise Amplifer

A CMOS dual-band low noise amplifer (LNA) design is presented.The purpose of th is work is intended to substitute only one LNA for two individual LNA's in dual -band transceivers for applications such as wireless local area network complying with both IEEE 802.11a and 802.11b/g.Dua l-band simultaneous input power and noise matching and load shaping are discuss ed.The chip is implemented in 0.25μm CMOS mixed and RF process.The measured pe rformance is summarized and discussed.

基于Dual-Mode MCMA+DD双模式盲均衡算法研究

为了兼顾误差性能和收敛速度,研究了将双模修正常量模算法和判决引导算法有机结合的双模式盲均衡算法.该算法在收敛初期采用双模修正常量模算法调整均衡器,然后根据判决条件自动切换到判决引导算法,最终实现信道的盲均衡.仿真结果表明:此算法收敛速度快、稳态误差小,并能在去除码间干扰的同时纠正信道固有的相位旋转,具有很好的实用性.