Biomedical Image Registration Research Articles

Deep-UEO: Empowering medical image registration with hybrid strategy based on deep learning and united equilibrium optimizer

Traditional optimization methods encounter challenges in determining the optimal geometric transformation parameters and often lacking to achieve global optimum which leads to extended computational time required for convergence. To address this challenge, a dual-phase strategy is proposed for multimodal biomedical image registration by amalgamating the united equilibrium optimizer (UEO) with Visual Geometry Group (VGG-19). In first phase, the input images are aligned by identifying the optimal values of rigid transformation parameters through UEO, with mutual information (MI) maximization serving as an objective function. In second phase, VGG19 is employed to enhance the reliability of resulting registered image obtained through UEO optimization for extracting both the low- and high-level features of reference and target images. Following this, dynamic inlier selection is employed to refine the matching process, integrating expectation maximization for iterative updating of inlier selection, thus improving registration accuracy. Finally, thin plate spline interpolation is utilized to calculate the transformation matrix, ensuring precise alignment between reference and UEO registered image. Extensive experiments are conducted on multimodal medical images sourced from publicly available repositories and clinical(in-house) datasets to validate the reliability of the proposed scheme. These experiments showcase substantial improvements in RMSE from 22.21094 to 10.983654, SSIM from 0.7190832 to 0.8409321, PSNR from 30.176044 to 34.095471, and CC from 0.9304633 to 0.9809432, outperforming state-of-the-art methods.

Review of Generative Adversarial Networks in mono- and cross-modal biomedical image registration.

Biomedical image registration refers to aligning corresponding anatomical structures among different images, which is critical to many tasks, such as brain atlas building, tumor growth monitoring, and image fusion-based medical diagnosis. However, high-throughput biomedical image registration remains challenging due to inherent variations in the intensity, texture, and anatomy resulting from different imaging modalities, different sample preparation methods, or different developmental stages of the imaged subject. Recently, Generative Adversarial Networks (GAN) have attracted increasing interest in both mono- and cross-modal biomedical image registrations due to their special ability to eliminate the modal variance and their adversarial training strategy. This paper provides a comprehensive survey of the GAN-based mono- and cross-modal biomedical image registration methods. According to the different implementation strategies, we organize the GAN-based mono- and cross-modal biomedical image registration methods into four categories: modality translation, symmetric learning, adversarial strategies, and joint training. The key concepts, the main contributions, and the advantages and disadvantages of the different strategies are summarized and discussed. Finally, we analyze the statistics of all the cited works from different points of view and reveal future trends for GAN-based biomedical image registration studies.

Progressive 3D biomedical image registration network based on deep self-calibration.

Three dimensional deformable image registration (DIR) is a key enabling technique in building digital neuronal atlases of the brain, which can model the local non-linear deformation between a pair of biomedical images and align the anatomical structures of different samples into one spatial coordinate system. And thus, the DIR is always conducted following a preprocessing of global linear registration to remove the large global deformations. However, imperfect preprocessing may leave some large non-linear deformations that cannot be handled well by existing DIR methods. The recently proposed cascaded registration network gives a primary solution to deal with such large non-linear deformations, but still suffers from loss of image details caused by continuous interpolation (information loss problem). In this article, a progressive image registration strategy based on deep self-calibration is proposed to deal with the large non-linear deformations without causing information loss and introducing additional parameters. More importantly, we also propose a novel hierarchical registration strategy to quickly achieve accurate multi-scale progressive registration. In addition, our method can implicitly and reasonably implement dynamic dataset augmentation. We have evaluated the proposed method on both optical and MRI image datasets with obtaining promising results, which demonstrate the superior performance of the proposed method over several other state-of-the-art approaches for deformable image registration.

Non-rigid registration of biomedical image for radiotherapy based on adaptive feature density flow

This paper presents a non-rigid biomedical image registration method based on Adaptive feature density flow (AFDF). Considering the muhiplicative characteristics of metadata signal for instance Magnetic Resonance Imaging (MRI) with high-resolution and high-sensitivity, the Non-Local Means Filtering Adapted to Rician Noise (NLMr) is employed to remove Rician noise rather than Gaussian noise with higher Peak Signal to Noise Ratio (PSNR). The deformation information and local shape information are homogeneous, so a flexible dimension of Fast Local Self-Similarity (FLSS) feature descriptor can be constructed to obtain adaptive FLSS (AFLSS) feature descriptor with adaptive Angle and Radial intervals. On the basis of AFLSS, the Adaptive feature density flow (AFDF) is proposed by calculating the dense feature flow field via Belief Propagation algorithm and thereby further optimized by the Pyramid iterative method. Finally, the image registration is completed by using the bicubic interpolation algorithm according to the dense feature flow field. Aiming at radiotherapy positioning control through precise robust registration of biomedical images, experiments of human brain MR images are carried out. The comparative results with PSNR, Root Mean Square Error (RMSE) and normalize mutual information (NMI) show that AFDF performs better than the optical flow method and Scale Invariant Feature Transform (SIFT) flow method, especially for the biomedical scenario with unconscious creep deformation.

Application of Image Registration Techniques to Medical CT and MR Images (Dept.E)

Image registration has become one of the most widely used techniques in computer vision. Its applications include optical flow, notion analysis, tracking, face detection, and biomedical image registration. In the present work, three different techniques of image registration were implemented and applied to both Computed Tomography (CT) and Magnetic Resonance (MR) images. The first technique is based on Cross Correlation (CC). The second approach depends on Control Points' Selection (CPS) from both the reference and the input images. The last technique is based on Maximization of Mutual Information (MMI) between We two images. The registrability is calculated for each image to measure its ability to provide unambiguous registration, by providing clear correlation peaks when registered with other subimage. Then, the three registration techniques were evaluated and compared using both the Weighted Peak Signal to Noise Ratio (WPSNR) and the Normalized Cross Correlation Coefficient (NCCC). The application of the selected techniques to CT and MR images has shown that registration based on MMI has given the best results and can be used efficiently for alignment of CT and MR images.

A full migration BBO algorithm with enhanced population quality bounds for multimodal biomedical image registration

Medical images acquired from different modalities give rise to many practical problems in image registration. Intensity-based registration techniques have been increasingly used in multimodal image registration; these techniques integrate different images that have shared content into a single representation, by transformation. The estimation of the optimal transformation requires the optimization of a similarity metric between the images. Recently, many optimization methods have been proposed that focus on the development of the optimization component. However, there is still room for large amounts of improvement, from both an efficiency point of view and a quality perspective. In this paper we present a new Biogeography-based Optimization (BBO) algorithm, the Biogeography-based Optimization algorithm with Elite Learning (BBO-EL), for multimodal medical image registration. First, we propose a hybrid full migration operator in which each individual has the chance to perform the migration operation and the whole population has the chance to expand the search space. In this way, the search ability of the BBO algorithm is enhanced and matches well the characteristics of multimodal medical image registration. In addition, considering that the quality of some individuals could be deteriorated as caused by the migration operation, we propose an undo operator on the deteriorated individuals. Thus, the lower bound of the whole population’s quality can be maintained at a higher level. Furthermore, in the original BBO algorithm, a number of good individuals might be not involved in the migration operation, and we present an elite learning operator that is based on social comparison theory to improve the upper bound of the whole population’s quality. Therefore, after improving both the lower bound and the upper bound of the whole population’s quality, the accuracy and the convergence speed of the multimodal medical registration can be greatly enhanced. The BBO-EL has been tested in many experiments on benchmark datasets include six kind of different modality images, from up to eighteen different patients, which can make up 54 multimodal registration scenarios. The BBO-EL obtained 30 best performance scenarios while the state-of-the-art algorithm obtained 21 scenarios. The results demonstrated that BBO-EL outperforms the state-of-the-art algorithm in most cases for practical problems.

Fast and Robust Symmetric Image Registration Based on Distances Combining Intensity and Spatial Information.

Intensity-based image registration approaches rely on similarity measures to guide the search for geometric correspondences with high affinity between images. The properties of the used measure are vital for the robustness and accuracy of the registration. In this study a symmetric, intensity interpolationfree, affine registration framework based on a combination of intensity and spatial information is proposed. The excellent performance of the framework is demonstrated on a combination of synthetic tests, recovering known transformations in the presence of noise, and real applications in biomedical and medical image registration, for both 2D and 3D images. The method exhibits greater robustness and higher accuracy than similarity measures in common use, when inserted into a standard gradientbased registration framework available as part of the open source Insight Segmentation and Registration Toolkit (ITK). The method is also empirically shown to have a low computational cost, making it practical for real applications. Source code is available.

Analysis of Machine Learning Algorithms for Diagnosis of Diffuse Lung Diseases.

Computational Intelligence Re-meets Medical Image Processing A Comparison of Some Nature-Inspired Optimization Metaheuristics Applied in Biomedical Image Registration Background Diffuse lung diseases (DLDs) are a diverse group of pulmonary disorders, characterized by inflammation of lung tissue, which may lead to permanent loss of the ability to breathe and death. Distinguishing among these diseases is challenging to physicians due their wide variety and unknown causes. Computer-aided diagnosis (CAD) is a useful approach to improve diagnostic accuracy, by combining information provided by experts with Machine Learning (ML) methods. Objectives Exploring the potential of dimensionality reduction combined with ML methods for diagnosis of DLDs; improving the classification accuracy over state-of-the-art methods. Methods A data set composed of 3252 regions of interest (ROIs) was used, from which 28 features were extracted per ROI. We used Principal Component Analysis, Linear Discriminant Analysis, and Stepwise Selection – Forward, Backward, and Forward-Backward to reduce feature dimensionality. The feature subsets obtained were used as input to the following ML methods: Support Vector Machine, Gaussian Mixture Model, k-Nearest Neighbor, and Deep Feedforward Neural Network. We also applied a Deep Convolutional Neural Network directly to the ROIs. Results We achieved the maximum reduction from 28 to 5 dimensions using LDA. The best classification results were obtained by DFNN, with 99.60% of overall accuracy. Conclusions This work contributes to the analysis and selection of features that can efficiently characterize the DLDs studied.

Biomedical Image Registration by means of Bacterial Foraging Paradigm

Image registration (IR) is the process of geometric overlaying or alignment f two or more 2D/3D images of the same scene (unimodal registration), taken r not at different time slots, from different angles, and/or by different image acquisition ystems (multimodal registration). Technically, image registration implies complex optimization of different parameters, performed at local or/and global evel. Local optimization methods often fail because functions of the involved metrics ith respect to transformation parameters are generally nonconvex and irregular, and lobal methods are required, at least at the beginning of the procedure. This paper resents a new evolutionary and bio-inspired robust approach for IR, Bacterial Foraging ptimization Algorithm (BFOA), which is adapted for PET-CT multimodal nd magnetic resonance image rigid registration. Results of optimizing the normalized utual information and normalized cross correlation similarity metrics validated he efficacy and precision of the proposed method by using a freely available medical mage database.

Deep Learning in Gastrointestinal Endoscopy.

Gastrointestinal (GI) endoscopy is used to inspect the lumen or interior of the GI tract for several purposes, including, (1) making a clinical diagnosis, in real time, based on the visual appearances; (2) taking targeted tissue samples for subsequent histopathological examination; and (3) in some cases, performing therapeutic interventions targeted at specific lesions. GI endoscopy is therefore predicated on the assumption that the operator-the endoscopist-is able to identify and characterize abnormalities or lesions accurately and reproducibly. However, as in other areas of clinical medicine, such as histopathology and radiology, many studies have documented marked interobserver and intraobserver variability in lesion recognition. Thus, there is a clear need and opportunity for techniques or methodologies that will enhance the quality of lesion recognition and diagnosis and improve the outcomes of GI endoscopy. Deep learning models provide a basis to make better clinical decisions in medical image analysis. Biomedical image segmentation, classification, and registration can be improved with deep learning. Recent evidence suggests that the application of deep learning methods to medical image analysis can contribute significantly to computer-aided diagnosis. Deep learning models are usually considered to be more flexible and provide reliable solutions for image analysis problems compared to conventional computer vision models. The use of fast computers offers the possibility of real-time support that is important for endoscopic diagnosis, which has to be made in real time. Advanced graphics processing units and cloud computing have also favored the use of machine learning, and more particularly, deep learning for patient care. This paper reviews the rapidly evolving literature on the feasibility of applying deep learning algorithms to endoscopic imaging.

Intensity Based Image Registration by Maximization of Mutual Information

Biomedical image registration, or geometric alignment of twodimensional and /or three-dimensional (3-D) image data, is becoming increasingly important in diagnosis, treatment planning, functional studies, and computer-guided therapies and in biomedical research [1]. Registration is an important problem and a fundamental task in image processing technique. In the medical image processing fields, some techniques are proposed to find a geometrical transformation that relates the points of an image to their corresponding points of another image. In recent years, multimodality image registration techniques are proposed in the medical imaging field. Especially, CT and MR imaging of the head for diagnosis and surgical planning indicates that physicians and surgeons gain important information from these modalities. In radiotherapy planning manual registration techniques performed on MR image and CT images of the brain. Now-adays, physicians segment the volume of interest (VOIs) from each set of slices manually. However, manual segmentation of the object area may require several hours for analysis. Furthermore, MDCT images and MR images contain more than 100 slices. Therefore, manual segmentation and registration method cannot apply for clinical application in the head CT and MR images. Many automatic and semiautomatic image registration methods have been proposed [2]. The main techniques of image registration are performed by the manual operation, using Landmark and using voxel information. In this paper, an automatic intensity based registration of head images by computer has been employed by applying maximization of mutual information. The primary objective of this paper is to increase accuracy of the registration and reduce the processing time. Experiments show our algorithm is a robust and efficient method which can yield accurate registration results.

A Microscopic Telepathology System for Multiresolution Computer-Aided Diagnostics

The aim of the presented system is simplification and speedup of the daily pathological examination routine. The system combines telepathology with computer-aided diagnostics algorithms. To the best of our knowledge, this is the first approach proposing such a comprehensive method. Our system is designed to accumulate knowledge through a learning process during diagnostics. Our system targets image acquisition and interpretation stages. The image acquisition subsystem solves various problems related to microscopical slide digitization such as biomedical image registration, data representation, and processing. The interpretation subsystem is based on Gabor filter texture features as well as on color features. A support vector machine classifier together with a feature selection is used for computer-aided diagnostics. The system design allows easy adaptation to a wide range of microscopical pathology examinations. The system is easily deployed and scaled. It has a low support cost and can aggregate a wide range of existing hardware. The experimental validation of the system is based on a database of more than three thousand samples. During the experimental evaluation, the system exhibited successful interaction with a pathologist.

An Approach to Multimodal Biomedical Image Registration Utilizing Particle Swarm Optimization

Biomedical image registration, or geometric alignment of two-dimensional and/or three-dimensional (3D) image data, is becoming increasingly important in diagnosis, treatment planning, functional studies, computer-guided therapies, and in biomedical research. Registration based on intensity values usually requires optimization of some similarity metric between the images. Local optimization techniques frequently fail because functions of these metrics with respect to transformation parameters are generally nonconvex and irregular and, therefore, global methods are often required. In this paper, a new evolutionary approach, particle swarm optimization, is adapted for single-slice 3D-to-3D biomedical image registration. A new hybrid particle swarm technique is proposed that incorporates initial user guidance. Multimodal registrations with initial orientations far from the ground truth were performed on three volumes from different modalities. Results of optimizing the normalized mutual information similarity metric were compared with various evolutionary strategies. The hybrid particle swarm technique produced more accurate registrations than the evolutionary strategies in many cases, with comparable convergence. These results demonstrate that particle swarm approaches, along with evolutionary techniques and local methods, are useful in image registration, and emphasize the need for hybrid approaches for difficult registration problems.