International Journal of Image and Graphics

Raycasting can display volumetric medical data in fine details and reveal crucial inner imaging information, while implicit surface is able to effectively model complex objects with high flexibility, combining these two rendering modalities together will provide comprehensive information of the scene and has wide applications in surgical simulation, image-guided intervention, and medical training. However, medical data rendering is based on texture depth at every sampling point, while mathematically modeled implicit surfaces do not have geometric information in texture space. It is a challenging task to visualize both physical scalar data and virtual implicit surfaces simultaneously. To address this issue, in this paper, we present a new dual-casting ray-based double modality data rendering algorithm and web-based software platform to visualize volumetric medical data and implicit surface in the same browser. The algorithm runs on graphics processing unit and casts two virtual rays from camera to each pixel on the display panel, where one ray travels through the mathematically defined scene for implicit surface rendering and the other one passes the 3D texture space for volumetric data visualization. The proposed algorithm can detect voxel depth information and algebraic surface models along each casting ray and dynamically enhance the visualized dual-modality data with the improved lighting model and transparency adjustment function. Moreover, auxiliary innovative techniques are also presented to enhance the shading and rendering features of interest. Our software platform can seamlessly visualize volumetric medical data and implicit surfaces in the same web browser over Internet.

This paper describes the Fast Radial Basis Function (RBF) method for cardiac motion tracking in 3D CT using non-rigid medical image registration based on parameterized (regular) surfaces. The technique is a point-based registration evaluation algorithm which does register 3D MR or CT images in real time. We first extract the surface of the whole heart 3D CT and its contrast enhanced part (left ventricle (LV) blood cavity) of each dataset with a semiautomatic contouring and a fully-automatic triangulation method followed by a global surface parameterization and optimization algorithm. In second step, a set of registration experiments are run to calculate the deformation field at various phases of cardiac motion or cycle from CT images, which results into significant deformation during each phase of a cycle. The surface points of the whole heart and LV are used to register the source systole image to various diastole target images taken at different phases during a heart beat. Our registration accuracy improves with the increase in number of salient feature points (i.e. optimized parameterized surfaces) and it has no effect on the speed of the algorithm (i.e. still less than a second). The results show that the target registration error is less than 3mm (2.53) and the performance of the Fast RBF algorithm is less than a second using a whole heart CT dataset of a single patient taken over the course of the entire cardiac cycle. At the end, the results for recovery (or analysis) of bigger deformation in heart CT images using the Fast RBF algorithm is compared to the state-of-the-art Free Form Deformation (FFD) registration technique. It is proved that the Fast RBF method is performing better in speed and slightly less accurate than the FFD (when measured in terms of NMI) due to iterative nature of the latter.

Automatic diagnosis of brain diseases based on brain connectivity network (BCN) classification is one of the hot research fields in medical image analysis. The functional brain network reflects the brain functional activities and structural brain network reflects the neural connections of the main brain regions. It is of great significance to explore and explain the inner mechanism of the brain and to understand and treat brain diseases. In this paper, based on the graph structure characteristics of brain network, the fusion model of functional brain network and structural brain network is designed to classify the diagnosis of brain mental diseases. Specifically, the main work of this paper is to use the Laplacian graph embed the information of diffusion tensor imaging, which contains the characteristics of structural brain networks, into the functional brain network with hyper-order functional connectivity information built based on functional magnetic resonance data using the sparse representation method, to obtain brain network with both functional and structural characteristics. Projection of the brain network and the two original modes data to the kernel space respectively and then classified by the multi-task learning method. Experiments on the epilepsy dataset show that our method has better performance than several state-of-the-art methods. In addition, brain regions and connections that are highly correlated with disease revealed by our method are discussed.

Conventional thresholding algorithms have had limited success with degraded document images. Recently, partial differential equations (PDEs) have been applied with good results. However, these are usually tailored to handle relatively few specific distortions. In this study, we combine an edge detection term with a linear binarization source term in a PDE formulation. Additionally, a new proposed diffusivity function further amplifies desired edges. It also suppresses undesired edges that comprise bleed-through effects. Furthermore, we develop the fractional variant of the proposed scheme, which further improves results and provides more flexibility. Moreover, nonlinear color spaces are utilized to improve binarization results for images with color distortion. The proposed scheme removes document image degradation such as bleed-through, stains, smudges, etc., and also restores faded text in the images. Experimental subjective and objective results show consistently superior performance of the proposed approach compared to the state-of-the-art PDE-based models.

Accurate delineation of the prostate in MR images is an essential step for treatment planning and volume estimation of the organ. Prostate segmentation is a challenging task due to its variable size and shape. Moreover, neighboring tissues have a low-contrast with the prostate. We propose a robust and precise automatic algorithm to define the prostate’s boundaries in MR images in this paper. First, we find the prostate’s ROI by a deep neural network and decrease the input image’s size. Next, a dynamic multi-atlas-based approach obtains the initial segmentation of the prostate. A watershed algorithm improves the initial segmentation at the next stage. Finally, an SSM algorithm keeps the result in the domain of allowable prostate shapes. The quantitative evaluation of 74 prostate volumes demonstrated that the proposed method yields a mean Dice coefficient of $0.83\pm 0.05$. In comparison with recent researches, our algorithm is robust against shape and size variations.

Image classification is a complicated process of classifying an image based on its visual representation. This paper portrays the need for adapting and applying a suitable image enhancement and denoising technique in order to arrive at a successful classification of data captured remotely. Biometric properties that are widely explored today are very important for authentication purposes. Noise may be the result of incorrect vein detection in the accepted image, thus explaining the need for a better development technique. This work provides subjective and objective analysis of the performance of various image enhancement filters in the spatial domain. After performing these pre-processing steps, the vein map and the corresponding vein graph can be easily obtained with minimal extraction steps, in which the appropriate Graph Matching method can be used to evaluate hand vein graphs thus performing the person authentication. The analysis result shows that the image enhancement filter performs better as an image enhancement filter compared to all other filters. Image quality measures (IQMs) are also tabulated for the evaluation of image quality.

This study proposes an image denoising algorithm based on sparse representation and Principal Component Analysis (PCA). The proposed algorithm includes the following steps. First, the noisy image is divided into overlapped $8\times 8$ blocks. Second, the discrete cosine transform is applied as a dictionary for the sparse representation of the vectors created by the overlapped blocks. To calculate the sparse vector, the orthogonal matching pursuit algorithm is used. Then, the dictionary is updated by means of the PCA algorithm to achieve the sparsest representation of vectors. Since the signal energy, unlike the noise energy, is concentrated on a small dataset by transforming into the PCA domain, the signal and noise can be well distinguished. The proposed algorithm was implemented in a MATLAB environment and its performance was evaluated on some standard grayscale images under different levels of standard deviations of white Gaussian noise by means of peak signal-to-noise ratio, structural similarity indexes, and visual effects. The experimental results demonstrate that the proposed denoising algorithm achieves significant improvement compared to dual-tree complex discrete wavelet transform and K-singular value decomposition image denoising methods. It also obtains competitive results with the block-matching and 3D filtering method, which is the current state-of-the-art for image denoising.

In this work, we aim to solve the multi-warping document image problems. We can overcome the limitations of the previous dewarping algorithms to recover the shape of the document. The proposed method is based on a well-defined pattern to simulate the distorted and undistorted connected component of document images. Some pairs of control points are selected for each connected component and its ground truth pattern to define the mapping function between them. The dewarping process transforms the warping connected component according to the geometric transformation defined by the calculated mapping function. Results on document dewarping dataset CBDAR demonstrate the effectiveness of our method. OCR error metrics are also used to evaluate the performance of the proposed approach.

This paper addresses the challenging task of word spotting in Arabic handwritten documents. We proposed a novel feature that we called Spatial Distribution of Ink at Keypoints (SDIK). The proposed feature captures the characteristics of Arabic handwriting concentrated at endpoints and branch points. SDIK feature quantizes the spatial repartition of ink pixels in the neighborhoods of keypoints. The resulting SDIK features are very fast to match, we take this advantage to match a query word with lines images rather than words images. By this matching mechanism, we overcome the hard task of segmenting an Arabic document into words. The method proposed in this study is tested on historical Arabic document with IBN SINA dataset and on modern handwriting with IFN/ENIT database. The obtained results are great of interest for retrieving query words in an Arabic document.

Images are widely used in engineering. Unfortunately, medical ultrasound images and synthetic aperture radar (SAR) images are mainly degraded by an intrinsic noise called speckle. Therefore, de-speckling is a main pre-processing stage for degraded images. In this paper, first, an optimized adaptive Wiener filter (OAWF) is proposed. OAWF can be applied to the input image without the need for logarithmic transform. In addition its performance is improved. Next, the coefficient of variation (CV) is computed from the input image. With the help of CV, the guided filter converts to an improved guided filter (IGF). Next, the improved guided filter is applied on the image. Subsequently, the fast bilateral filter is applied on the image. The proposed filter has a better image detail preservation compared to some other standard methods. The experimental outcomes show that the proposed denoising algorithm is able to preserve image details and edges compared with other de-speckling methods.

One of the most often used techniques to represent color images is quaternion algebra. This study introduces the quaternion Krawtchouk moments, QKrMs, as a new set of moments to represent color images. Krawtchouk moments (KrMs) represent one type of discrete moments. QKrMs use traditional Krawtchouk moments of each color channel to describe color images. This new set of moments is defined by using orthogonal polynomials called the Krawtchouk polynomials. The stability against the translation, rotation, and scaling transformations for QKrMs is discussed. The performance of the proposed QKrMs is evaluated against other discrete quaternion moments for image reconstruction capability, toughness against various types of noise, invariance to similarity transformations, color face image recognition, and CPU elapsed times.

Magnetic resonance imaging (MRI) is a dynamic and safe imaging technique in medical imaging. Recently, parallel MRI (pMRI) is widely used for accelerating conventional MRI. Both frequency and image domain-based reconstructions are the most attractive methods for generating the image from multi-channel k-space data. Compressed sensing (CS) is a recently used procedure to reduce the acquisition time of conventional MRI. This reduction is achieved by taking fewer measurements from the fully sampled k-space data. Therefore, applying the CS technique in pMRI is the most emerging way for further improving the acquisition time that is a tremendous research interest. However, as the phase encoding plane may be perpendicular or parallel to the coil elements plane, finding the exact domain for CS in pMRI reconstruction is a major challenging issue. In this work, the application of the CS technique in pMRI in both domains is investigated. Later some widely used methodologies are presented as the nonlinear reconstruction algorithm of CS in pMRI. Finally, a discussion is performed based on CS in pMRI to perceive the reality of different reconstruction algorithms at a glance for finding preferred methodologies.

The coronavirus disease (COVID-19) pandemic that is caused by the SARS-CoV2 has spread all over the world. It is an infectious disease that can spread from person to person. The severity of the disease can be categorized into five categories namely asymptomatic, mild, moderate, severe, and critical. From the reported cases thus, it has been seen that 80% of the cases that test positive with COVID-19 infection have less than moderate complications, whereas 20% of the positive cases develop severe and critical complications. The virus infects the lungs of an individual, therefore, it has been observed that the X-ray and computed tomography (CT) scan images of the infected people can be used by the machine learning-based application programs to predict the presence of the infection. Therefore, in the proposed work, a Convolutional Neural Network model based upon the DenseNet architecture is being used to predict the presence of COVID-19 infection using the CT scan images of the chest. The proposed work has been carried out using the dataset of the CT images from the COVID CT Dataset. It has 349 images marked as COVID-19 positive and 397 images have been marked as COVID-19 negative. The proposed system can categorize the test set images with an accuracy of 91.4%. The proposed method is capable of detecting the presence of COVID-19 infection with good accuracy using the chest CT scan images of the humans.

The field of hexagonal image processing is concerned with the creation of image processing systems that combine the advantages of biological model-based evolutionary motivated frameworks. The structure and functionality of artificial neural networks were inspired by biological processes. The fundamental framework of recording and output devices limits their present state of the art. Prior neural networks have used square or hexagonal style input to completely connected layers, which resulted in a high coherence problem between two adjacent hexagonal kernel layers due to pooling. Previous research does not design the self-data structure to support convolution to increase computational efficiency, so it violates the convolution and pooling operator, which greatly degrades the image process performance. This paper introduces a novel paradigm Proficient Deep Learning-based Hexrep Neural Network that overcomes major significant problems in image operations structure constraint, coherence problem, and violation of convolution and pooling operator and achieves hexagonal image processing.