基于双域Transformer耦合特征学习的CT截断数据重建模型

目的为解决CT扫描视野(FOV)不足导致的截断伪影和图像结构失真问题,本文提出了一种基于投影和图像双域Transformer耦合特征学习的CT截断数据重建模型(DDTrans)。方法基于Transformer网络分别构建投影域和图像域恢复模型,利用Transformer注意力模块的远距离依赖建模能力捕捉全局结构特征来恢复投影数据信息,增强重建图像。在投影域和图像域网络之间构建可微Radon反投影算子层,使得DDTrans能够进行端到端训练。此外,引入投影一致性损失来约束图像前投影结果,进一步提升图像重建的准确性。结果Mayo仿真数据实验结果表明,在部分截断和内扫描两种截断情况下,本文方法DDTrans在去除FOV边缘的截断伪影和恢复FOV外部信息等方面效果均优于对比算法。结论DDTrans模型可以有效去除CT截断伪影,确保FOV内数据的精确重建,同时实现FOV外部数据的近似重建。

he adaptive chirplet transform and its application in GPR target detection

GPR has become an important geophysical method in UXO and landmine detection, for it can detect both metal and non-metallic targets. However, it is difficult to remove the strong clutters from surface-layer reflection and soil due to the low signal to noise ratio of GPR data. In this paper, we use the adaptive chirplet transform to reject these clutters based on their character and then pick up the signal from the UXO by the transform based on the Radon-Wigner distribution. The results from the processing show that the clutter can be rejected effectively and the target response can be measured with high SNR.

基于Transformer的肺肿瘤三维CT图像分割

基于信息学技术自动分割病人的肺部CT图像,有助于医生对于肺癌患者的早期诊断,提取和整合图像区域间的空间关联,对于提升肺肿瘤分割性能是十分重要的。本文提出了一个新的基于Transformer的分割模型,用于肺肿瘤三维CT图像分割、学习和整合此类关联。本文分别设计了带有混合多头图像区域节点注意力的Transformer模块和类别注意力模块,学习并融合了肺部CT图像的空间层面和通道层面的信息。将新的基于Transformer的分割模型同其他较为先进的模型进行了对比实验,实验结果表明新的模型在骰子系数、交并比和豪斯多夫距离等方面优于其他模型。

MCIF-Transformer Mask RCNN:Multi-Branch Cross-Scale Interactive Feature Fusion Transformer Model for PET/CT Lung Tumor Instance Segmentation

The precise detection and segmentation of tumor lesions are very important for lung cancer computer-aided diagnosis.However,in PET/CT(Positron Emission Tomography/Computed Tomography)lung images,the lesion shapes are complex,the edges are blurred,and the sample numbers are unbalanced.To solve these problems,this paper proposes a Multi-branch Cross-scale Interactive Feature fusion Transformer model(MCIF-Transformer Mask RCNN)for PET/CT lung tumor instance segmentation,The main innovative works of this paper are as follows:Firstly,the ResNet-Transformer backbone network is used to extract global feature and local feature in lung images.The pixel dependence relationship is established in local and non-local fields to improve the model perception ability.Secondly,the Cross-scale Interactive Feature Enhancement auxiliary network is designed to provide the shallow features to the deep features,and the cross-scale interactive feature enhancement module(CIFEM)is used to enhance the attention ability of the fine-grained features.Thirdly,the Cross-scale Interactive Feature fusion FPN network(CIF-FPN)is constructed to realize bidirectional interactive fusion between deep features and shallow features,and the low-level features are enhanced in deep semantic features.Finally,4 ablation experiments,3 comparison experiments of detection,3 comparison experiments of segmentation and 6 comparison experiments with two-stage and single-stage instance segmentation networks are done on PET/CT lung medical image datasets.The results showed that APdet,APseg,ARdet and ARseg indexes are improved by 5.5%,5.15%,3.11%and 6.79%compared with Mask RCNN(resnet50).Based on the above research,the precise detection and segmentation of the lesion region are realized in this paper.This method has positive significance for the detection of lung tumors.

Roll angular rate extraction based on modified spline-kernelled chirplet transform

The roll angular rate is much crucial for the guidance and control of the projectile.Yet the high-speed rotation of the projectile brings severe challenges to the direct measurement of the roll angular rate.Nevertheless,the radial magnetometer signal is modulated by the high-speed rotation,thus the roll angular rate can be achieved by extracting the instantaneous frequency of the radial magnetometer signal.The objective of this study is to find out a precise instantaneous frequency extraction method to obtain an accurate roll angular rate.To reach this goal,a modified spline-kernelled chirplet transform(MSCT)algorithm is proposed in this paper.Due to the nonlinear frequency modulation characteristics of the radial magnetometer signal,the existing time-frequency analysis methods in literature cannot obtain an excellent energy concentration in the time-frequency plane,thereby leading to a terrible instantaneous frequency extraction accuracy.However,the MSCT can overcome the problem of bad energy concentration by replacing the short-time Fourier transform operator with the Chirp Z-transform operator based on the original spline-kernelled chirplet transform.The introduction of Chirp Z-transform can improve the construction accuracy of the transform kernel.Since the construction accuracy of the transform kernel determines the concentration of time-frequency distribution,the MSCT can obtain a more precise instantaneous frequency.The performance of the MSCT was evaluated by a series of numerical simulations,high-speed turntable experiments,and real flight tests.The evaluation results show that the MSCT has an excellent ability to process the nonlinear frequency modulation signal,and can accurately extract the roll angular rate for the high spinning projectiles.

Segmentation of Head and Neck Tumors Using Dual PET/CT Imaging:Comparative Analysis of 2D,2.5D,and 3D Approaches Using UNet Transformer

The segmentation of head and neck(H&N)tumors in dual Positron Emission Tomography/Computed Tomogra-phy(PET/CT)imaging is a critical task in medical imaging,providing essential information for diagnosis,treatment planning,and outcome prediction.Motivated by the need for more accurate and robust segmentation methods,this study addresses key research gaps in the application of deep learning techniques to multimodal medical images.Specifically,it investigates the limitations of existing 2D and 3D models in capturing complex tumor structures and proposes an innovative 2.5D UNet Transformer model as a solution.The primary research questions guiding this study are:(1)How can the integration of convolutional neural networks(CNNs)and transformer networks enhance segmentation accuracy in dual PET/CT imaging?(2)What are the comparative advantages of 2D,2.5D,and 3D model configurations in this context?To answer these questions,we aimed to develop and evaluate advanced deep-learning models that leverage the strengths of both CNNs and transformers.Our proposed methodology involved a comprehensive preprocessing pipeline,including normalization,contrast enhancement,and resampling,followed by segmentation using 2D,2.5D,and 3D UNet Transformer models.The models were trained and tested on three diverse datasets:HeckTor2022,AutoPET2023,and SegRap2023.Performance was assessed using metrics such as Dice Similarity Coefficient,Jaccard Index,Average Surface Distance(ASD),and Relative Absolute Volume Difference(RAVD).The findings demonstrate that the 2.5D UNet Transformer model consistently outperformed the 2D and 3D models across most metrics,achieving the highest Dice and Jaccard values,indicating superior segmentation accuracy.For instance,on the HeckTor2022 dataset,the 2.5D model achieved a Dice score of 81.777 and a Jaccard index of 0.705,surpassing other model configurations.The 3D model showed strong boundary delineation performance but exhibited variability across datasets,while the 2D model,although effective,generally underperformed compared to its 2.5D and 3D counterparts.Compared to related literature,our study confirms the advantages of incorporating additional spatial context,as seen in the improved performance of the 2.5D model.This research fills a significant gap by providing a detailed comparative analysis of different model dimensions and their impact on H&N segmentation accuracy in dual PET/CT imaging.

An Improved Chirplet Transform and Its Application for Harmonics Detection

The chirplet transform is the generalization form of fast Fourier transform , short-time Fourier transform, and wavelet transform. It has the most flexible time frequency window and successfully used in practices. However, the chirplet transform has not inherent inverse transform, and can not overcome the signal reconstructing problem. In this paper, we proposed the improved chirplet transform (ICT) and constructed the inverse ICT. Finally, by simulating the harmonic voltages, The power of the improved chirplet transform are illustrated for harmonic detection. The contours clearly showed the harmonic occurrence time and harmonic duration.

基于增强CT图像和Swin Transformer网络的食管癌T分期智能诊断模型的构建与评估

目的基于增强CT图像和Swin Transformer网络,拟构建食管癌T分期智能诊断模型。方法收集2018年1月至2022年4月在陆军军医大学第一附属医院和山西省肿瘤医院胸外科经病理证实为食管癌的150例患者的45000张术前增强CT图像。经过UperNet Swin网络自动分割和肿瘤体积的计算,使用ResNet50、Swin Transformer和VIT 3个网络进行食管癌T分期智能诊断模型的构建。使用精准率、召回率、F1-score、特异度以及阴性预测值(negative predictive value,NPV)等指标在150例内部数据集上评价模型性能,描绘混淆矩阵和ROC曲线。结果在3个食管癌T分期诊断的模型中,Swin Transformer模型结合肿瘤体积、病理信息等特征的分期诊断效果最好,T1~T4期的精准率分别为1.00、0.67、0.83、1.00,AUC为0.861,优于ResNet50和VIT分期诊断模型,它们的精准率分别为0.13、0.27、0.59、0.81和0.03、0.14、0.56、0.75,AUC分别是0.611和0.542。结论与ResNet50和VIT网络比较,Swin Transformer网络能够更精准进行食管癌智能T分期诊断。

面向肺炎CT图像识别的DL-CTNet模型

肺炎常缺乏明显呼吸系症状,症状多不典型,易发生漏诊、错诊.利用深度学习技术辅助医务人员安全、高效地检测感染者是一种有效途径.针对COVID-19感染者CT图像的磨玻璃影、铺路石征、血管扩张等特点,提出一种可有效地提取CT图像中的局部与全局特征的轻量级模型——DL-CTNet.输入预处理的CT图像后,首先采用空洞卷积和动态双路径多尺度特征融合(D-DMFF)模块的2个支路提取浅层特征;然后使用局部与全局特征拼接模块(LGFC)中的D-DMFF模块提取局部特征、Swin Transformer提取全局特征,并通过拼接获得深层特征;最后经过全连接层输出分类标签.实验结果表明,在2个CT图像数据集上,验证了LGFC模块以及DL-CTNet的低复杂度与有效性;DL-CTNet的分类准确率高达98.613%,与其他方法相比,其能更准确地识别肺炎的CT图像.

一种新的术中X线与术前CT图像配准方法

目的本研究旨在配准胸主动脉血管内修复术(thoracic endovascular aortic repair,TEVAR)术中X线与术前CT图像,为TEVAR支架植入提供精确安全的导航。然而,现有配准算法存在无法有效弥合投影CT图像生成的数字重建影像(digitally reconstructed radiography,DRR)与X线图像之间的域间差异和难以获得图像分割标签的问题。因此,需要提出新的方法来改善这一问题。方法本文提出了一种新的配准框架,该框架结合了基于生成对抗网络(generative adversarial network,GAN)的域自适应网络和基于Transformer的配准网络。基于GAN的域自适应网络将X线图像的风格迁移到DRR图像上,使两者在图像风格上更接近。基于Transformer的配准网络采用CNN与跨模态变换器(cross-modality transformer,CMT)相结合的模式,直接配准X线与CT图像,无需进行图像分割。结果本文在208对标定的TEVAR术中X线与CT图像对上对新的配准方法进行了验证。与其他域适应方法相比,本文所采用的CycleGAN网络作为风格转换模块,有效减小了DRR图像与X线图像之间的域间差异。消融实验结果进一步证实,配准网络中的全局局部感知模块(global-local perception module,GLPM)对提高配准精度具有明显作用,而空间缩减(spatial reduction,SR)则有效缩短了配准时间。通过对比现有方法和本文方法在真实患者X线与CT图像对上的配准效果,本文的方法在配准精度和成功率方面均表现出最佳性能。结论本文提出的新的X线与CT图像配准方法有效克服了现有方法存在的域间差异以及难以获得分割标签的问题。

基于多项式Chirplet变换的线性调频引信干扰波形设计方法

针对现代战场环境下线性调频引信干扰技术的迫切需求,提出多项式Chirplet变换的方法实现低信噪比下对调频连续波无线电引信的信号参数估计,并针对调频连续波引信重构了干扰信号。基于线性小波变换原理,选取适当的核函数,将无线电引信信号经过旋转与平移操作后进行短时傅里叶变换,再利用遗传算法求得变换核的最优参数,实现对调频连续波信号的参数估计,并重构干扰信号。仿真对比结果表明,该方法可以在较低信噪比下实现对调频连续波无线电引信信号载波频率、调制周期、最大频偏等参数进行精准估计,相较于周期调制类干扰,重构干扰信号能够以更低功率对调频引信产生较好的干扰效果。

基于Transformer的脊柱CT图像分割

脊柱侧弯是青少年群体常见的一种脊柱疾病,老年群体中因脊柱骨质疏松引起的脊柱骨折也尤为普遍。CT成像技术作为脊柱外科的主要检查手段之一,广泛用于临床以及研究目的的筛查,诊断和图像引导治疗。研究以脊柱CT图像为研究对象,将目前在NLP领域表现优异的Transformer模型与经典的U-Net图像分割网络相结合,运用到CT图像的分割处理工作当中;同时在模型训练过程中基于脊柱自身的结构特点,采用由粗到精的训练方法,首先对脊柱的各个椎骨进行定位模型训练,然后在定位结果的基础上再训练分割模型。最终模型的分割结果与真实值之间的Dice相似系数达到了94.37%以上,实验结果表明了该方法的有效性以及临床应用的可行性。

多期相CT合成辅助的腹部多器官图像分割

目的提出并探讨使用多期相CT合成辅助腹部多器官分割方法。方法提出多期相CT合成辅助腹部多器官分割,多期相CT能够充分提供同一器官不同的图像细节,从而为分割模型提供充分的全面的语义信息,提升腹部多个器官分割的性能。提出基于多头自注意力感知的多期相CT合成方法,引入基于多头自注意力机制的Transformer模块,提升合成网络捕捉长距离语义信息的能力,扩大网络的感受野,并且引入感知损失,在特征层面对合成图像与真实图像特征之间的差异最小化,与Transformer模块有协同作用,从而合成出更清晰、更高质量的多期相CT图像。结果使用南方医院的多期相CT数据集训练模型。其中用526例多期相CT训练合成模型,利用动脉期增强动脉CT(A.CECT)合成出平扫CT(NECT)、静脉期CECT(V.CECT)、延迟期CECT(D.CECT)的平均最大化绝对误差(MAE)分别为19.192±3.381、20.140±2.676、22.538±2.874,结合统计学对比,本文方法优于对比的其他图像合成方法(P<0.05)。多期相CT合成辅助的腹部多器官分割方法验证在内部验证集上进行验证平均Dice系数(DSC)为0.847,在外部验证集上进行验证平均DSC为0.823。结论本文方法能够合成出高质量的多期相CT图像以有效缓解不同期相CT之间存在的配准无法解决的误差问题,同时提高腹部13器官的分割性能,具有良好的泛化性能。

基于Transformer和CNN的低剂量CT图像去噪网络

低剂量计算机断层扫描(Low-dose Computed Tomography, LDCT)在临床中有着广泛的应用,可以有效减轻对病人的辐射剂量。但是成像后的低剂量CT图像中含有明显的噪声和条形伪影,影响医师的诊断。提出了一种基于Transformer和CNN的去噪网络,该网络是一种改进的编解码网络架构,其编码端的每一层由卷积模块与Transformer模块融合而成,用来提取每一层的局部特征和全局特征,同时引入融合模块用来有效地融合提取的局部特征和全局特征。并把融合后的特征通过跳跃连接融入解码端对应的层,解码端的每一层通过卷积模块提取有效特征进而重建去噪后的图像。在真实数据集Mayo上的实验结果说明所提出的网络不仅可以有效去除噪声,还能够保持图像的边缘。

Predictive Potential of FDG-PET/CT for Histological Transformation in Patients with Indolent Lymphoma

Backgrounds and Purpose—In indolent non-Hodgkin lymphoma, histological transformation is a dramatic event which reduces the prognosis significantly. SUVmax values from FDG-PET/CT help differentiate between aggressive and indolent lymphomas, and transformed indolent lymphomas also show an increased FDG uptake. Possibly FDG uptake increases early in the clinical course and could predict histological transformation. Our objective was to predict histological transformation in indolent lymphomas from initial staging FDG-PET/CT. Patients and Methods—A retrospective study was performed. Patients with biopsy-proven indolent lymphoma who had had initial staging FDG-PET/CT were included. Qualitative (foci compared with FDG uptake liver) and semiquantitative (SUVmax-value per focus) analyses were performed of all abnormal foci. Patient characteristics and outcome were evaluated. Results—We included 88 patients, 5 of whom developed a histological transformation. Semiquantitative analysis showed a relation between maximum standardized uptake value and histological transformation (odds ratio 1.25, 95% CI 1.024 - 1.513). Qualitative analysis showed a negative predictive relation of FDG uptake less than or equal to liver in the occurrence of histological transformation. Transformation-free survival was 100% over 30 months in those with FDG uptake lower than or equal to liver. More FDG uptake than liver showed transformation-free survival of 88% over 30 months. Conclusion—Qualitative analysis of staging FDG-PET/CT in indolent lymphomas could be useful to rule out transformation in the next 30 months. In our study, semiquantitative analysis was statistically significantly associated with histological transformation and maximum standardized uptake value. However, because of the small number of patients, cautious interpretation of the results is warranted. More studies are needed to investigate the role of staging PET/CT in patient with indolent non-Hodkin lymphoma in the prediction of transformation.

基于DCIF-GAN的肺部肿瘤PET/CT跨模态医学图像融合

基于生成对抗网络(Generative Adversarial Network,GAN)的医学图像融合是计算机辅助诊断领域的研究热点之一,但是现有基于GAN的融合方法存在训练不稳定,提取图像的局部和全局上下文语义信息能力不足,交互融合程度不够等问题。针对上述问题,本文提出了双耦合交互式融合GAN(Dual-Coupled Interactive Fusion GAN,DCIFGAN)。首先,设计了双生成器双鉴别器GAN,通过权值共享机制实现生成器之间和鉴别器之间的耦合,通过全局自注意力机制实现交互式融合;第二,设计耦合CNN-Transformer的特征提取模块(Coupled CNN-Transformer Feature Extraction Module,CC-TFEM)和特征重构模块(CNN-Transformer Feature Reconstruction Module,C-TFRM),提升了对同一模态图像内部的局部和全局特征信息提取能力;第三,设计跨模态交互式融合模块(Cross Model Intermodal Fusion Module,CMIFM),通过跨模态自注意力机制,进一步整合不同模态间的全局交互信息。为了验证本文模型的有效性,在肺部肿瘤PET/CT医学图像数据集上进行实验,该文方法在平均梯度,空间频率,结构相似度,标准差,峰值信噪比,信息熵等上与其他四种方法中最优方法相比,分别提高了1.38%,0.39%,29.05%,30.23%,0.18%,4.63%。模型能够突出病变区域信息,融合图像结构清晰且纹理细节丰富。

Research on the Reason for Transformer Differential Protection Mal-operation Caused by Sympathetic Inrush

The paper is concerned with whether CT transient saturation, which caused by sympathetic inrush, is the direct reason for mal-operation of transformer differential relays. In order to analyze transforming characteristics of saturated CT, the gain of each harmonic current transforming from primary winding to secondary winding of transient saturation CT is calculated. Based on the fact that series sympathetic inrush is easier to lead differential relay mal-operation than parallel sympathetic inrush, the effect of CT saturation on differential relay?in the case of both parallel sympathetic inrush and series sympathetic inrush are investigated respectively. Theory analysis and simulation show that higher harmonic will be transferred by saturated CT more easily than lower harmonic, and then the second harmonic proportion of CT secondary side is higher than that of CT primary side. Thus CT transient saturation itself is not essential reason for differential relay mal-operation. Parallel sympathetic inrush and series sympathetic inrush will lead to different location CT saturated. Parallel sympathetic inrush will lead to CT located system side saturated;second harmonic ratio of differential current increase and dead angle is large. Series sympathetic inrush will lead to CT located load side saturated;second harmonic ratio of differential current decrease and dead angle is small.

基于特征融合的低剂量CT图像降噪方法

近年来低剂量CT(Low Dose CT,LDCT)被广泛应用于临床诊断中,但LDCT会产生不规则的噪声。已有的降噪方法往往缺乏对全局特征信息的考虑,以及不注重边缘特征信息和重建图像的视觉效果。为此,提出了一种基于特征融合的低剂量CT图像降噪方法。首先,利用Transformer优异的全局感受野提取图像的全局特征信息,并利用卷积神经网络(Convolutional Neural Network,CNN)良好的局部特征提取能力提取图像的局部特征信息。在Transformer模块中加入维度变换思想,以更好地抑制噪声;在CNN模块中使用稠密连接的方式将浅层网络的特征信息复用于深层网络中,以此保存更多的特征信息。其次,为了获取更加丰富的图像细节特征,使用了改进的索伯边缘增强算子来加强模型对边缘特征信息的提取能力。最后,将Transformer模块和CNN模块获取的特征信息进行融合并输出重建图像。此外,为了使降噪重建后的图像有更好的质量和视觉效果,设计了一个多尺度复合损失函数。实验表明:在AAPM-Mayo数据集的降噪实验中,与当前主流的LDCT图像降噪方法相比,本文方法取得了更好的降噪效果。

基于改进样条Chirplet变换的结构瞬时频率识别

为了提高结构瞬时频率识别的精度,对样条Chirplet变换进行改进,并结合能量集中度原则优化窗函数的参数.利用解析信号和两层框架模型验证改进样条Chirplet变换识别瞬时频率的准确性,再结合试验验证改进样条Chirplet变换能有效识别实际工程结构的瞬时频率.

用于低剂量CT图像去噪的多级双树复小波网络

基于卷积神经网络(CNN)的图像去噪方法能有效去除低剂量计算机断层扫描(CT)图像伴随的伪影和噪声,从而确保CT设备输出高质量图像同时降低辐射,这对患者健康和医学诊断具有重要意义。为了进一步提高低剂量CT图像的质量,提出一种小波域去噪网络MDTNet。首先,基于双树复小波变换(DTCWT)构造多级编解码去噪网络,在多个尺度上提取特征以保留更多高频细节;然后,利用扩展的像素重排技术替代卷积上下采样,实现多级输入和特征融合,从而降低计算复杂度;最后,通过大量训练找到最佳的去噪模型,即二级MDTNet配合LeGall滤波器和Qshift_b滤波器,并选择较大尺寸的CT图像作为训练数据。使用AAPM数据集评估MDTNet的性能,实验结果表明,MDTNet能有效去除条纹状伪影和噪声,在定量和定性评估中性能均优于同类型去噪方法。与FWDNet相比,对于1 mm的切片,MDTNet的平均峰值信噪比(PSNR)和结构相似性指数(SSIM)分别提高了0.0887 dB和0.0024;对于3 mm的切片,分别提升了0.1443 dB和0.003。对于单张512×512像素的低剂量CT图像去噪,MDTNet在GPU上仅需0.193 s。MDTNet在保持高效率的同时保留了更多的高频细节,能够为低剂量CT图像去噪提供一种新的框架。