血管网光声成像的机器学习抗散射仿真研究

乳腺癌已成为全球女性发病率最高的肿瘤疾病,微血管成像对乳腺癌的治疗方案和预后有重要意义。光声层析成像术(Photoacoustic Tomography, PAT)可有效对乳腺癌内微血管网进行成像,但肿瘤组织内部的异质微结构和钙化点的散射对成像质量影响较大。针对该问题,文章基于U-Net的卷积神经网络对不同颗粒散射条件下软组织中血管网图像散斑开展仿真研究。仿真结果表明,该神经网络可以学习光声散斑图像和成像目标之间的映射关系,提取出隐藏在噪声中的血管光声信号,并重建出轮廓清晰、背景清晰的高质量血管图像,表明U-Net网络可以从高度模糊的散射图像中提取出有效的光声信息,实现目标图像的高清重建,在乳腺癌的诊断成像中具有广阔的应用前景。     

采用稀疏测量数据的有限角度光声层析成像的研究进展

生物光声层析(Photoacoustic Tomography,PAT)成像可以反映生物组织的光吸收分布,定量测量组织的光吸收系数和散射系数,进而分析组织成分,为疾病的早期诊断和治疗提供可靠的依据。由于成像目标特殊的几何结构以及成像装置的机械结构、空间位置和成像时间等的限制,超声探测器只能在有限的角度范围内扫描,采集到稀疏的光声测量数据,导致重建图像中出现伪影和失真。针对有限角度扫描和稀疏测量数据问题,对目前主流的光声图像重建算法进行综述和分析。     

基于阵列探测方式的时域光声成像系统

将阵列超声探头和超声相控技术与光声成像相结合的成像系统，与采用水听器的单探头旋转扫描光声成像系统相比，避免了机械旋转机构给光声信号采集所带来的不稳定性，提高了数据采集速度．时域光声信号由64阵元线阵超声探头以电子相控聚焦的方式进行线性扫描采集，然后通过时域后向投影算法进行光声图像的重建．采用波长532nm、重复频率10Hz的脉冲激光，系统可快速重建样品内部光学吸收分部的二维图像，单帧图像数据采集时间小于200s，成像横向分辨率小于2mm．实验结果表明，采用此方法可显著提高系统对光声信号的扫描稳定性和成像效率，该系统是一种有潜在临床应用价值的光声成像系统．     

基于生成对抗网络的CT图像无监督超分辨率分析

提高计算机断层成像(CT)医疗影像的分辨率有助于医生更精确地识别病变部位,具有重要临床诊断意义。本文研究在没有高-低分辨率图像对数据的条件下,使用仅包含低分辨率图像的数据集,通过降质网络和注入噪声获得与真实图像同域的低分辨率图像,进而构造接近天然图像对的训练数据集。并且设计了包括超分辨生成器、超分辨鉴别器和超分辨特征提取器的超分辨率生成对抗网络(DeSRGAN),实现对CT影像4倍超分辨率分析。实验测试表明,超分辨率分析生成的4倍CT图像在NIQE、BRISQUE和PIQE等无参考图像质量评估指标的定量对比中,DeSRGAN方法均优于最新的单图像超分辨率的增强型深度残差网络(EDSR)、残差信道注意力网络(RCAN)、增强型超分辨率生成对抗性网络(ESRGAN)等方法生成的图像。同时在直观视觉效果上,DeSRGAN方法生成的图像具有更清晰细节和更好感知效果。     

多光谱光声层析技术在肿瘤学中的应用进展

多光谱光声层析技术是一种新兴的光声成像技术,兼具光声断层扫描和多光谱成像的优点。该成像技术需要利用内源性和外源性光声成像对比剂进行成像。成像时,激光发射器发射多个波长的激光束照射组织,组织发生热弹性膨胀并产生超声波,将超声换能器接收的光声信号进行后处理,分解光谱信息并重建图像。目前,多光谱光声层析技术已广泛用于多种肿瘤的研究。文章着重介绍光声成像对比剂和多光谱光声层析技术的发展近况以及在临床转化中取得的研究进展。     

生物声学成像中声速不均匀性解决方法的研究进展

对于以超声波为载体的生物医学声学成像(如超声、光声和磁声成像等)技术,为了简化问题,常在假设待测组织内声速恒定的前提下,重建组织内的声阻抗、光吸收分布、光学特性参数分布或者电导率分布等。但是,实际生物组织内部的声速是存在差异的(最大可达10%),因而在此假设前提下重建出的图像通常是不准确的。在介绍声速不均匀性对声学图像重建影响的基础上,对超声、光声和磁声成像中解决声速不均匀问题的主要方法,特别是光声层析成像中重建组织内声速分布的主要方法进行总结和归纳,讨论各自的优点和不足,并展望未来的可能发展方向。     

基于通道注意力与迁移学习的红外图像超分辨率重建算法

针对现有红外图像分辨率低、质量不高的问题,提出了基于通道注意力与迁移学习的红外图像超分辨率重建方法.该方法设计了一个深度卷积神经网络,融入通道注意力机制来增强网络的学习能力,并且使用残差学习方式来减轻梯度爆炸或消失问题,加速网络的收敛.考虑到高质量的红外图像难以采集、数目不足的情况,将网络的训练分成两步:第一步使用自然...     

光声成像技术在生物医学中的研究进展

光声成像是近年来发展较快的无损检测技术,其高分辨率、高对比度的特点使其成为生物医学检测技术的主要发展方向之一。文中从光声成像系统、光声探测器和图像重建算法的角度,对光声成像技术进行了分析。在此基础上,分别结合光声计算层析成像、光声显微成像和光声内窥成像对光声成像系统进行了阐述,探讨了光声探测器在多探头、阵列式、MEMS微型化和光纤F-P腔等方面的研究进展,比较了典型图像重建算法的应用特点,并指出了基于光声技术的生物医学无损成像系统的高分辨率、大探测深度、实时性强、小型化、低成本的研究方向。     

基于BP神经网络和遗传算法的电阻抗图像重建算法

医学电阻抗成像(MEIT)技术是一个复杂的非线性问题,图像重建算法对所构图像质量具有重要的影响。由于传统的BP神经网络算法存在训练速度慢且易陷于局部极小等特点,采用全局寻优的遗传算法(GA)实现BP网络连接权的学习,并将该混合算法用于电阻抗断层图像(EIT)重建。实验结果表明,该算法所构图像无论被测场域中心还是边缘处均具有较高的图像分辨率,与典型的反投影及灵敏度系数算法所构图像相比,其质量得到明显的改善。     

磁感应磁声电导率图像重建研究

0引言磁感应磁声成像(Magneto-acoustic Tomography with Magnetic Induction,MAT-MI)是一种基于磁、声、电多物理场耦合的生物组织电导率成像新方法,由于其同时具有电阻抗成像高对比度和超声成像高空间分辨率的优点,所以有着很高的研究价值和广阔的应用前景[1-5]。在以往的MAT-MI研究中,常用无指向性的点声源模型对由洛伦兹力引起的组织振动进行研究,     

Angular motion point spread function model considering aberrations and defocus effects

When motion blur is considered, the optics point spread function (PSF) is conventionally assumed to be fixed, and therefore cascading of the motion optical transfer function (OTF) with the optics OTF is allowed. However, in angular motion conditions, the image is distorted by space-variant effects of wavefront aberrations, defocus, and motion blur. The proposed model considers these effects and formulates a combined space-variant PSF obtained from the angle-dependent optics PSF and the motion PSF that acts as a weighting function. Results of comparison of the new angular-motion-dependent PSF and the traditional PSF show significant differences. To simplify the proposed model, an efficient approximation is suggested and evaluated.     

Compensating for nonstationary blurring by further blurring and deconvolution

In many imaging systems, the point spread function (PSF) is nonstationary. Usually, a computation‐intensive iterative algorithm is used to deblur the nonstationary PSF. This article presents a new idea of using a noniterative method to compensate for the spatially variant PSF. This method first further blurs the image with a nonstationary kernel so that the resultant image has a stationary PSF, then deblurs the resultant image using an efficient decovolution technique. The proposed method is illustrated and implemented by single photon emission computed tomography applications. © 2009 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 19, 221–226, 2009     

Accurate retrieval of region of interest for estimating point spread function and image deblurring

Extraction of an optimal region of interest (ROI) is crucial in many image processing applications, such as estimation of the point spread function (PSF) and blind deconvolution (BD). Although the amount of publications on PSF and BD is quite extensive; however, the work on ROI estimation has not received much attention. Existing methods which used heuristic models are not only time-consuming but also computationally expensive. In this paper, we proposed a new ROI retrieval scheme based on image partitioning and entropy measurement feedback. This method has low computation cost since it contains no matrix operations. Comprehensive experiments on real and synthetic datasets revealed that the proposed method is competitive when compared with existing search techniques, averaging at 26.1?dB, 0.46 and 1.44 on peak signal-to-noise ratio, universal image quality index and error ratio scales, respectively. On average, the proposed method takes less than 10?s to retrieve the ROI which is significantly faster compared to established solution.     

MLEM deconvolution of protein X-ray diffraction images based on a multiple-PSF model

In this paper we analyze the degradation of protein X-ray diffraction images by diffuse light distortion (DLD). In order to correct the degradation, a new multiple point spread function (PSF) model is introduced and used to restore X-ray diffraction image data (XRD). Raw PSFs are collected from isolated spots in high-resolution areas on the diffraction patterns which represent the orientation of DLDs. An adaptive ridge regression (ARR) technique is used to remove noise from the raw PSF data. A target Gaussian function is used to model the raw PSFs. A maximum likelihood expectation maximization (MLEM) algorithm combined with a multi-PSF model is employed to restore high intensity, asymmetrical protein X-ray diffraction data. Experimental results using a single and multiple PSFs are presented and discussed. We show that using a multiple PSF model in the deconvolution algorithm improved the quality of the XRD and as a result the spot integration error (/spl chi//sup 2/) and corresponding electron density map are improved.     

Generation of a high depth of focus with constant transversal spot size using a phase-only pupil filter

A novel approach is proposed to obtain an extended depth of focus (DoF) in white light with constant transversal spot size within the DoF. It combines a phase-only pupil filter based on multiplexed radial zones with alternating quartic phase functions. The design is first tested via numerical simulations of the point spread function (PSF) based on the scalar diffraction theory. The results for a fourfold gain of the depth of focus are experimentally verified with a phase-only spatial light modulator liquid crystal device combined with a 3D PSF measurement system. A close conformity between the experimental and simulation results proves the effectiveness of the proposed approach.     

基于超像素的联合能量主动轮廓CT图像分割方法

为解决医学CT图像主动轮廓分割方法中对初始轮廓敏感的问题,提出一种基于超像素和卷积神经网络的人体器官CT图像联合能量函数主动轮廓分割方法。该方法首先基于超像素分割对CT图像进行超像素网格化,并通过卷积神经网络进行超像素分类确定边缘超像素;然后提取边缘超像素的种子点组成初始轮廓;最后在提取的初始轮廓基础上,通过求解本文提出的综合能量函数最小值实现人体器官分割。实验结果表明,本文方法与先进的U-Net方法相比平均Dice系数提高5%,为临床CT图像病变诊断提供理论基础和新的解决方案。     

Research on Multi-View Image Reconstruction Technology Based on Auto-Encoding Learning

Traditional three-dimensional (3D) image reconstruction method, which highly dependent on the environment and has poor reconstruction effect, is easy to lead to mismatch and poor real-time performance. The accuracy of feature extraction from multiple images affects the reliability and real-time performance of 3D reconstruction technology. To solve the problem, a multi-view image 3D reconstruction algorithm based on self-encoding convolutional neural network is proposed in this paper. The algorithm first extracts the feature information of multiple two-dimensional (2D) images based on scale and rotation invariance parameters of Scale-invariant feature transform (SIFT) operator. Secondly, self-encoding learning neural network is introduced into the feature refinement process to take full advantage of its feature extraction ability. Then, Fish-Net is used to replace the U-Net structure inside the self-encoding network to improve gradient propagation between U-Net structures, and Generative Adversarial Networks (GAN) loss function is used to replace mean square error (MSE) to better express image features, discarding useless features to obtain effective image features. Finally, an incremental structure from motion (SFM) algorithm is performed to calculate rotation matrix and translation vector of the camera, and the feature points are triangulated to obtain a sparse spatial point cloud, and meshlab software is used to display the results. Simulation experiments show that compared with the traditional method, the image feature extraction method proposed in this paper can significantly improve the rendering effect of 3D point cloud, with an accuracy rate of 92.5% and a reconstruction complete rate of 83.6%.     

Supergridded cone-beam reconstruction and its application to point-spread function calculation

In cone-beam computed tomography (CBCT), the volumetric reconstruction may in principle assume an arbitrarily fine grid. The supergridded cone-beam reconstruction refers to reconstructing the object domain or a subvolume thereof with a grid that is finer than the proper computed tomography sampling grid (as determined by gantry geometry and detector discreteness). This technique can naturally reduce the voxelization effect, thereby retaining more details for object reproduction. The grid refinement is usually limited to two or three refinement levels because the detail pursuit is eventually limited by the detector discreteness. The volume reconstruction is usually targeted to a local volume of interest due to the cubic growth in a three-dimensional (3D) array size. As an application, we used this technique for 3D point-spread function (PSF) measurement of a CBCT system by reconstructing edge spread profiles in a refined grid. Through an experiment with a Teflon ball on a CBCT system, we demonstrated the supergridded volume reconstruction (based on a Feldcamp algorithm) and the CBCT PSF measurement (based on an edge-blurring technique). In comparison with a postreconstruction image refinement technique (upsampling and interpolation), the supergridded reconstruction could produce better PSFs (in terms of a smaller FWHM and PSF fitting error).     

Ultrasonic imaging using a computed point spread function

An explicit point spread function (PSF) evaluator in the frequency domain is described for an ultrasonic transducer operating in the pulse-echo mode. The PSF evaluator employs the patch element model for transducer field determination and scattered field assessment from a small but finite "point" reflector. The PSF for a planar transducer in a medium has been evaluated in the near and the far field. The computed PSFs were used to deconvolve and restore surface images, obtained experimentally, of a single hole and a five-hole cluster in an Al calibration block. A calibration plot is arrived at for estimating, without the need for deconvolution, the actual diameters of circular reflectors from apparent diameters obtained experimentally for a single-medium imaging configuration. The PSF, when the transducer and the point reflector are in two media separated by a planar interface, was evaluated in the near and far field. The computed PSFs were used to deconvolve and restore subsurface images, obtained experimentally, of flat bottom holes (FBHs) in an Al calibration block. We show that the PSF, in the presence of a planar interface, can be obtained from a single-medium PSF model using an effective single-medium path length concept. The PSFs and modulation transfer functions (MTFs) are evaluated for spherical focused and annular transducers and compared with those for the planar transducer. We identify imaging distances to get better-resolved images when using planar, spherical focused, and annular transducers.     

Research on the magnetic resonance imaging brain tumor segmentation algorithm based on DO-UNet

With the social and economic development and the improvement of people's living standards, smart medical care is booming, and medical image processing is becoming more and more popular in research, of which brain tumor segmentation is an important branch of medical image processing. However, the manual segmentation method of brain tumors requires a lot of time and effort from the doctor and has a great impact on the treatment of patients. In order to solve this problem, we propose a DO-UNet model for magnetic resonance imaging brain tumor image segmentation based on attention mechanism and multi-scale feature fusion to realize fully automatic segmentation of brain tumors. Firstly, we replace the convolution blocks in the original U-Net model with the residual modules to prevent the gradient disappearing. Secondly, the multi-scale feature fusion is added to the skip connection of U-Net to fuse the low-level features and high-level features more effectively. In addition, in the decoding stage, we add an attention mechanism to increase the weight of effective information and avoid information redundancy. Finally, we replace the traditional convolution in the model with DO-Conv to speed up the network training and improve the segmentation accuracy. In order to evaluate the model, we used the BraTS2018, BraTS2019, and BraTS2020 datasets to train the improved model and validate it online, respectively. Experimental results show that the DO-UNet model can effectively improve the accuracy of brain tumor segmentation and has good segmentation performance.