基于TransUnet和残差结构的脑肿瘤图像分割方法

目的 卷积神经网络由于具有在捕捉长距离依赖能上的缺陷，其限制了分割模型的进一步提升^[1]。针对以往医学图像分割网络中卷积的感受野太小，对小型医学图像数据集训练参数非常多的深度网络容易过拟合问题，基于TransUnet网络模型结合ViT与Unet的特性对模型进行优化。方法 利用TransUnet与ResNet残差结构与注意力机制结合，在底层CNN模块引入注意力机制提取特征，引入残差结构加深算法，使用卷积操作代替混合编码器中的最大池化减少特征丢失，增大了卷积中的感受野，提高了脑胶质瘤图像的分割精度。结果 在BraTS比赛的数据集中对脑胶质瘤的分割Dice达到95.22%,较原TransUnet模型提升了2%。结论 研究基于TransUnet优化算法的脑肿瘤图像分割方法鲁棒性强。对现有的两种神经网络方法进行对比，均低于本文改进模型，表明对于医学图像的分割精度有提升。     

基于CT图像的汽车碰撞假人上衣模型点云数据提取研究

CT辅助逆向工程技术是在不破坏实物原型基础上,利用CT扫描获得物体内外表面有序点云数据,是现代产品设计与建模的重要手段。对于具有薄壁结构的产品,传统方法在利用其CT断层图像提取点云数据时,会存在轮廓线交错、断点等问题,影响着点云的提取精度。本文针对汽车碰撞假人上衣这种薄壁件表面点云数据提取困难的问题,提出一种适用于复杂薄壁件内外点云提取的新方法。实验证明,该方法能够得到高精度的复杂薄壁件内外表面点云数据,为后续的CAD建模提供基础。     

基于线性滤波的脑CT图像颅内结构的提取

针对现有的提取颅脑CT图像颅腔内结构算法自动性较差的问题,提出了一种新算法。该算法首先根据颅脑颅骨组织宽度大、灰度值高的特点,利用线性空间滤波方法提取颅脑颅骨的粗略轮廓;然后利用图像数学形态学方法和基于阅值的水平左右扫描算法,实现颅内组织的自动化分割;最后,通过对100例颅脑CT检查病例的图像进行实验,来验证算法的可行性。实验结果表明,该算法使99％具有完整颅骨环的颅腔内结构实现计算机自动化分割,分割结果准确。     

双能DR物质识别算法在CT成像系统中的应用

本文介绍一种适用于计算机断层成像（CT）的双能透视（DR）物质识别算法，此算法基于单能CT重建图像，通过对CT图像的分割及对断层几何信息的提取，分块重建材料有效原子序数及电子密度的分布。结合各种扫描轨迹的CT成像系统，可以实现有效的物质材料识别。同时，对比于传统的双能CT方法，本方法结合单能CT重建图像，改善了双能DR识别效果，能够实现较为精确的物质识别，在安全检查等应用领域有着现实意义。     

基于Kinect的物体表面重建方法研究

随着测量技术的发展,基于深度图像的物体表面重建技术得到了越来越多的应用。本文实现了一种基于Kinect的物体表面重建方法。通过Kinect采集原始数据并经过感兴趣剔除、原始点云滤波、不同帧点云之间的配准、物体表面重建以及网格平滑等步骤,构造出了物体表面的模型。与传统的基于深度图像的重建方法相比,本文方法简便实用,能够获得较好的重建效果。     

超高分辨率地形场景纹理图像生成方法研究

目的 航空倾斜摄影图像的空间分辨率受限于无人机飞行高度和图像传感器精度,导致三维重建得到的三维模型的纹理分辨率低,清晰度差.可利用超分辨率重建算法还原图像的纹理细节并提高图像的空间分辨率,最终获得具有超高分辨率纹理的三维模型.方法 利用单帧超分辨率重建算法LapSRN处理航空倾斜摄影图像,可以得到的高分辨率图像,并进一...     

基于霍夫变换的射波刀等中心误差胶片分析算法

目的 提出一种基于霍夫变换(HT)的新算法提高射波刀自动质量保证(AQA)测试胶片影像分析的准确性与稳定性,并探讨胶片图像的扫描分辨率对测试结果的影响。方法 获取9对胶片对AQA模体进行分析测试,首先利用中值滤波对灰度化后的胶片影像预处理,去除噪声干扰;再使用全局阈值对图像进行二值化分割,对分割后的图像进行边缘检测并利用HT提取胶片边缘直线,将胶片图像变换至正确位置;最后利用边缘检测和HT提取出影像中射野投影圆和钨球投影圆的圆心,通过分析同心度最终得到AQA测试的误差分析结果。结果 所有样本分别使用本算法与原软件算法所得等中心误差结果差异无统计学意义(P>0.05),其标准差差异有统计学意义(P=0.027);表明本算法在保证了胶片分析准确性的同时具有更好的稳定性。两种方法均不能通过采用分辨率更高的胶片影像来提高分析结果的准确性和稳定性。结论 本算法排除了胶片扫描摆位误差造成的干扰,为射波刀系统AQA工作提供了一种更稳定的途径。     

CT辅助逆向工程技术在畸齿矫正中的应用研究

如何快速、高精度获取齿形CAD模型是目前畸齿矫正行业中的技术难题。介绍了一种基于3D-CT的牙颌数字模型快速获取方法。利用3D-CT扫描获得牙齿石膏阳模精细的三维灰度图像;对三维CT图像进行相应的图像、图形处理,获取实物内外表面点云数据;并将其输入CAD设计平台,逆求出实物原型的CAD模型。与传统逆求方法相比,该方法具有扫描速度快、点云数据各向等分辨率和逆求精度高的优点。     

斑点噪声分布拟合的乳腺超声病灶分割方法

乳腺超声图像常存在斑点噪声,给自动分割带来了一定干扰。为了提升图像质量而进行的去噪操作可能导致病灶边界信息的损失。本文探索将噪声处理融合到分割中而提出新方法,采用统计特征概率分布形式描述乘性斑点噪声分布并融入水平集模型,从而将图像去噪与目标分割结合成一个统一体,实现此类型噪声分布图像的高效自动分割。本文方法对人工合成图像分割得到了较好的效果,对真实乳腺超声图像病灶的分割测试实现了速度提升,在分割精度方法则显示出分化结果,通过分析原因指出探索性的改进措施。     

基于多特征提取的遥感图像机场目标自动检测

传统的遥感图像机场跑道自动目标检测由于仅提取灰度特征常产生过分割现象,本文采用灰度特征和纹理特征相结合的方法进一步提高跑道的检测精度。利用阈值对遥感图像进行初始分割,以定位感兴趣区域（ROI）,再利用EM算法估计ROI区域训练样本,引入马尔可夫随机场（MRF）模型,分割机场跑道。实验表明MRF可以很好地描述空间连续性,可以达到精确检测机场跑道的目的。     

HT系统兆伏级CT图像质量和剂量计算研究

目的 定量对HT系统兆伏级CT图像质量及剂量计算准确性进行评估。方法 应用兆伏级CT扫描Cheese phantom模体,分别对图像成像几何精度、噪声、图像均匀性、空间分辨率、密度-CT值的转化及剂量计算准确性进行测试,并与常规千伏级CT进行对比分析。结果 兆伏级CT图像的成像几何精度在三维方向均<2 mm,MVCT图像噪声、均匀性、空间分辨率都差于千伏级CT。基于CT图像进行剂量重建的DVH均与千伏级CT计划的DVH有很好一致性。结论 兆伏级CT图像几何尺寸精确,成像剂量低,剂量计算精确,满足于临床要求。     

Reconstruction of Breast Using A Laser Lithographic Model: A Case Report

A three-dimensional laser scanner enables the measurement of surface data from objects with no contact, both quickly and in three dimensions. A 40-year-old woman with a right breast cancer had undergone a simple mastectomy. The surface of the chest was then measured with the non-contact three-dimensional laser scanner. Using the mirror image of data obtained by the same system, a mask-like model of the breast was made using laser lithography technology. This model was used for reconstruction of the breast after mastectomy. It was effective in determining the position and form of the breast which was to be repaired.     

Compressed sensing for reduction of noise and artefacts in direct PET image reconstruction

AimImage reconstruction in positron emission tomography (PET) can be performed using either direct or iterative methods. Direct reconstruction methods need a short reconstruction time. However, for data containing few counts, they often result in poor visual images with high noise and reconstruction artefacts. Iterative reconstruction methods such as ordered subset expectation maximization (OSEM) can lead to overestimation of activity in cold regions distorting quantitative analysis. The present work investigates the possibilities to reduce noise and reconstruction artefacts of direct reconstruction methods using compressed sensing (CS).Materials and methodsRaw data are generated either using Monte Carlo simulations using GATE or are taken from PET measurements with a Siemens Inveon small-animal PET scanner. The fully sampled dataset was reconstructed using filtered backprojection (FBP) and reduced in Fourier space by multiplication with an incoherently undersampled sampling pattern, followed by an additional reconstruction with CS. Different sampling patterns are used and an average of the reconstructions is taken. The images are compared to the results of an OSEM reconstruction and quantified using signal-to-noise ratio (SNR).ResultsThe application of the proposed CS post-processing technique clearly improves the image contrast. Dependent on the undersampling factor, noise and artefacts are reduced resulting in an SNR that is increased up to 3.4-fold. For short acquisition times with low count statistics the SNR of the CS reconstructed image exceeds the SNR of the OSEM reconstruction.ConclusionEspecially for low count data, the proposed CS-based post-processing method applied to FBP reconstructed PET images enhances the image quality significantly.     

Cone-beam-CT guided radiation therapy: technical implementation.

BACKGROUND AND PURPOSE: X-ray volumetric imaging system (XVI) mounted on a linear accelerator is available for image guidance applications. In preparation for clinical implementation, phantom and patient imaging studies were conducted to determine the irradiation parameters that would trade-off image quality, patient dose and scanning time. PATIENTS AND METHODS: The XVI image quality and imaging dose were benchmarked against those obtained with a helical CT scanner for a head and body phantom. The irradiation parameters were varied including the total imaging dose, number of projections, field of view, reconstruction resolution and use of a scatter rejection grid. We characterized the image quality based on relative contrast, noise, contrast to noise ratio (CNR) and point spread function (PSF). XVI scans of pelvis, head and neck and lung patients were acquired and submitted to a range of observers to identify the favorable reconstruction parameters. RESULTS: Phantom studies have demonstrated that a scatter rejection grid reduces photon scattering and improves the image uniformity. For the body phantom, the helical CT and the wide field XVI technique produce similar image quality, with surface doses of 0.025 and 0.044 Gy respectively. We have demonstrated that the local tomography technique improves the image contrast and the CNR while reducing the skin dose by 40-50% compared to the wide field technique. Clinical scans of head and neck, lung and prostate patients present good soft tissue contrast and excellent bone definition. CONCLUSIONS: With adjustment of irradiation parameters and an imaging surface dose of less than 0.05 Gy, high quality XVI images can be obtained for a phantom simulating the body thickness. XVI is currently feasible for image-guided treatments of head and neck, torso and pelvic areas using soft tissue and bony structures.     

一种基于聚类的消失点自动测量方法

在人工环境中有很多平行的直线和相互正交的平面。本文提出了一种对相机拍摄的人工环境中的中远距离场景图像,自动获取其空间中三个方向相互正交的消失点的方法。利用中远距离场景中的平行直线在图像中倾角接近这一特点对图像中的边缘线段进行聚类初始化,并利用空间中平行直线在图像中相交于消失点这一事实进行聚类。三个消失点的三角形标准以及相机焦距标准将有助于消除虚假消失点。实验结果表明,该方法对中远距离场景的消失点检测可行、快速。     

Parameters Influencing Volume and Activity Quantitation in Spect

Several factors influence the accuracy to which information on the in vivo activity concentration can be obtained from SPECT images. The accuracy in image contrast and the quantitation of volume and activity has here been examined for a SPECT system aimed for absorbed dose determinations in systemic radiation therapy. The influence of source dimensions, object contrast, noise level, energy window width and the reconstruction filter was examined. All the investigated parameters affected the recoveries. Volumes of above 20 cm³ could be quantified to within ±20% for object contrasts of above 96%, and within ±40% for object contrasts above 81%, providing limited levels of noise. Both the image contrast and the volume quantification is expected to be improved in situations with low object contrasts, if accurate correction for scattered photons can be developed. Obtained limitations in the quantitation of activity with the present method should be reduced by development of a more accurate background subtraction technique     

On-line implant reconstruction in HDR brachytherapy.

BACKGROUND AND PURPOSE: To evaluate the accuracy of on-line planning in an Integrated Brachytherapy Unit (IBU) using dedicated image distortion correction algorithms, correcting the geometric distortion and magnetic distortion separately, and to determine the effect of the reconstruction accuracy on clinical treatment plans in terms of deviations in treatment time and dose. PATIENTS AND METHODS: The reconstruction accuracy has been measured using 20 markers, positioned at well known locations in a QA phantom. Treatment plans of two phantoms representing clinical implant geometries, have been compared with reference plans to determine the effect of the reconstruction accuracy on the treatment plan. Before clinical introduction, treatment plans of three representative patients, based on on-line reconstruction, have been compared with reference plans. RESULTS: The average reconstruction error for 10 in. images reduces from -0.6 mm (range -2.6 to +1.0 mm) to -0.2 mm (range -1.2 to +0.6 mm) after image distortion correction and for 15 in. images from 0.8 mm (range -0.5 to +3.0 mm) to 0.0 mm (range -0.8 to +0.8 mm). The error in case of eccentric positioning of the phantom, i.e. 0.8 mm (range -1.0 to +3.3 mm), reduces to 0.1 mm (range -0.5 to +0.9 mm). Correction of the image distortions reduces the deviation in the calculated treatment time of maximally 2.7% to less than 0.8% in case of eccentrically positioned clinical phantoms. The deviation in the treatment time or reference dose in the plans based on on-line reconstruction with image distortion correction of the three patient examples is smaller than 0.3%. CONCLUSIONS: Accurate on-line implant reconstruction using the IBU localiser and dedicated correction algorithms separating the geometric distortion and the magnetic distortion is possible. The results fulfill the minimum requirements as imposed by the Netherlands Commission on Radiation Dosimetry (NCS) without limitations regarding the usable range of the field of view of the image intensifier. However, the C-arm angle is limited to those angles for which magnetic distortion corrections have been obtained.     

激光聚变中存在吸收的图像重建

本文提出一种处理激光等离子体中含吸收的图像重建方法 ,运用最大熵原理推导出适用于含吸收的重建算法。采用不同的图像实例进行模拟计算 ,计算结果表明采用吸收校正方法的精度明显高于不考虑吸收校正的精度。     

CT扫描/重建参数对三维治疗计划系统影像的影响

[目的]研究CT模拟定位中,CT扫描/重建参数对三维治疗计划重建的三维假体的几何精确度的影响.[方法]在西门子CT模拟机（Somatom plus 4）上扫描自制模体,扫描所得图像登记到ADAC三维治疗计划系统重建成三维假体,测量假体的相关坐标数据并与模体的实际数据相比较;对Catphan 412模体扫描并测量各组图像的实际层厚,讨论实际层厚的变化对计划系统中登记影像的几何精度影响.[结果]（1）CT扫描所采用的不同扫描/重建参数对三维计划系统中重建的三维假体的左右及上下方向的几何精度影响不大,但对重建假体的前后方向（即模体扫描的步进方向）的几何精度有一定的影响.（2）CT扫描所采用的螺距及重建模式会对层厚敏感度曲线（SSP）半高宽值产生影响,该变化对重建假体的前后方向几何精度同样有一定的影响.[结论]重建CT图像的前后方向的几何误差是随着扫描层厚增加而增加,主要是由于CT扫描的部分容积效应影响.单纯增加螺距或使用360度线性内插（Wide）重建模式,都会引起CT图像实际层厚的增加,引起更大的容积效应影响.同时部分容积效应也会导致三维治疗计划系统中数字重建影像（DRR）分辨率的降低.     

Measurement of MRI scanner noise

The present paper describes a simple method for the analysis of MRI scanner noise. Besides the heating of body tissue by strong RF radiation and the formation of circular currents in the body induced bey switching field gradients, a noise level of more than 100 dB(A) during the measurement belongs to the potential risks of MRI. This risk is of particular concern for staff and accompanying persons who remain close to the scanner for different reasons (e. g., monitoring of anesthetized patients, reassuring of children). For this reason, and given the scanty information on noise provided in the manuals of the scanners, it is useful to quantify the noise level more exactly. This applies also to the evaluation of different sound-reducing methods for the patient. This presents the results of noise level measurements in the tomograph and in its surrounding, with and without noise reduction by headphones.