Intermediate Slices Research Articles

A novel GAN-based three-axis mutually supervised super-resolution reconstruction method for rectal cancer MR image

Background and objectiveThis study aims to enhance the resolution in the axial direction of rectal cancer magnetic resonance (MR) imaging scans to improve the accuracy of visual interpretation and quantitative analysis. MR imaging is a critical technique for the diagnosis and treatment planning of rectal cancer. However, obtaining high-resolution MR images is both time-consuming and costly. As a result, many hospitals store only a limited number of slices, often leading to low-resolution MR images, particularly in the axial plane. Given the importance of image resolution in accurate assessment, these low-resolution images frequently lack the necessary detail, posing substantial challenges for both human experts and computer-aided diagnostic systems. Image super-resolution (SR), a technique developed to enhance image resolution, was originally applied to natural images. Its success has since led to its application in various other tasks, especially in the reconstruction of low-resolution MR images. However, most existing SR methods fail to account for all anatomical planes during reconstruction, leading to unsatisfactory results when applied to rectal cancer MR images. MethodsIn this paper, we propose a GAN-based three-axis mutually supervised super-resolution reconstruction method tailored for low-resolution rectal cancer MR images. Our approach involves performing one-dimensional (1D) intra-slice SR reconstruction along the axial direction for both the sagittal and coronal planes, coupled with inter-slice SR reconstruction based on slice synthesis in the axial direction. To further enhance the accuracy of super-resolution reconstruction, we introduce a consistency supervision mechanism across the reconstruction results of different axes, promoting mutual learning between each axis. A key innovation of our method is the introduction of Depth-GAN for synthesize intermediate slices in the axial plane, incorporating depth information and leveraging Generative Adversarial Networks (GANs) for this purpose. Additionally, we enhance the accuracy of intermediate slice synthesis by employing a combination of supervised and unsupervised interactive learning techniques throughout the process. ResultsWe conducted extensive ablation studies and comparative analyses with existing methods to validate the effectiveness of our approach. On the test set from Shanxi Cancer Hospital, our method achieved a Peak Signal-to-Noise Ratio (PSNR) of 34.62 and a Structural Similarity Index (SSIM) of 96.34 %. These promising results demonstrate the superiority of our method.

Generation of synthetic intermediate slices in 3D OCT cubes for improving pathology detection and monitoring

OCT is a non-invasive imaging technique commonly used to obtain 3D volumes of the ocular structure. These volumes allow the monitoring of ocular and systemic diseases through the observation of subtle changes in the different structures present in the eye. In order to observe these changes it is essential that the OCT volumes have a high resolution in all axes, but unfortunately there is an inverse relationship between the quality of the OCT images and the number of slices of the cube. This results in routine clinical examinations using cubes that generally contain high-resolution images with few slices. This lack of slices complicates the monitoring of changes in the retina hindering the diagnostic process and reducing the effectiveness of 3D visualizations. Therefore, increasing the cross-sectional resolution of OCT cubes would improve the visualization of these changes aiding the clinician in the diagnostic process. In this work we present a novel fully automatic methodology to perform the synthesis of intermediate slices of OCT image volumes in an unsupervised manner. To perform this synthesis, we propose a fully convolutional neural network architecture that uses information from two adjacent slices to generate the intermediate synthetic slice. We also propose a training methodology, where we use three adjacent slices to train the network by contrastive learning and image reconstruction. We test our methodology with three different types of OCT volumes commonly used in the clinical setting and validate the quality of the synthetic slices created with several medical experts and using an expert system.

Pathology Synthesis of 3D-Consistent Cardiac MR Images using 2D VAEs and GANs

We propose a method for synthesizing cardiac magnetic resonance (MR) images with plausible heart pathologies and realistic appearances for the purpose of generating labeled data for the application of supervised deep-learning (DL) training. The image synthesis consists of label deformation and label-to-image translation tasks. The former is achieved via latent space interpolation in a VAE model, while the latter is accomplished via a label-conditional GAN model. We devise three approaches for label manipulation in the latent space of the trained VAE model; i) intra-subject synthesis aiming to interpolate the intermediate slices of a subject to increase the through-plane resolution, ii) inter-subject synthesis aiming to interpolate the geometry and appearance of intermediate images between two dissimilar subjects acquired with different scanner vendors, and iii) pathology synthesis aiming to synthesize a series of pseudo-pathological synthetic subjects with characteristics of a desired heart disease. Furthermore, we propose to model the relationship between 2D slices in the latent space of the VAE prior to reconstruction for generating 3D-consistent subjects from stacking up 2D slice-by-slice generations. We demonstrate that such an approach could provide a solution to diversify and enrich an available database of cardiac MR images and to pave the way for the development of generalizable DL-based image analysis algorithms. We quantitatively evaluate the quality of the synthesized data in an augmentation scenario to achieve generalization and robustness to multi-vendor and multi-disease data for image segmentation. Our code is available at <a href='https://github.com/sinaamirrajab/CardiacPathologySynthesis'>https://github.com/sinaamirrajab/CardiacPathologySynthesis</a>

Medical Image Enhancement Based on Volumetric Tissue Segmentation Fusion (Uni-Stable 3D Method)

The 3D Uni-stable is a new method for 3D medical image enhancement which produces 3D Images of high contrast from the scanned anisotropic scaling images. This is done by estimating some intermediate slices through resizing the original scans. Rescaling has been achieved at three different levels: rescaling of eigenvalues of diffusion, rescaling the Scalar Indexes from the original eigenvalues, and rescaling the cluster maps of the segmentation of the original Scalar Indexes. Four interpolation methods have been employed at each level and four clustering algorithms have been employed in the process. The 3D Uni-stable image is almost universal as it combines variety of algorithms points of views into one 3D probability map. This reduces boundary-overlapping among different tissues, and hence improves the uniqueness of the segmentation problem solution. The stability factor of the 3D Uni-stable-Images is measured by maximum match analysis between the cluster maps which are generated from 3D Uni-stable images using variety of clustering methods with respect to true fact references for 5 different brains and the resultant standard deviations of Uni-stable images maximum match analysis in both threshold and tissue to brain ratio are much lower than Mean Diffusivity and Fractional Anisotropy scalar indexes for both CSF/non-CSF and WM/non-WM respectively

Inter-Slice Resolution Improvement Using Convolutional Neural Network with Orbital Bone Edge-Aware in Facial CT Images.

The 3D modeling of orbital bones in facial CT images is essential to provide a customized implant for reconstructions of orbit and related structures during surgery. However, 3D models of the orbital bone show an aliasing effect and disconnected thin bone in the inter-slice direction because the slice thickness is two to three times larger than the pixel spacing. To improve the inter-slice resolution of facial CT images, we propose a method based on a 2D convolutional neural network (CNN) that uses the spatial information on the sagittal and axial planes and the orbital bone edge-aware (OBE) loss. First, intermediate slices are generated on the sagittal plane. Second, the generated intermediate slices are transformed to an axial image, which is then compared with the original axial image. To generate intermediate slices with an accurate orbital bone structure, the OBE loss considering the orbital bone structure on the sagittal and axial planes is used. To improve the perceptual quality of the generated intermediate slices, the feature map difference loss is additionally used on the axial plane. In the experiment, the proposed method showed the best performance among bilinear and bicubic interpolations, 3D SRGAN, and a 2D CNN-based method. Experimental results confirmed that the proposed method can generate intermediate slices with clear edges of thin bones as well as cortical bones on both the sagittal and the axial plane.

Weighted 3D volume reconstruction from series of slice data using a modified Allen–Cahn equation

In this study, we develop a fast and accurate computational method for a weighted three-dimensional (3D) volume reconstruction from a series of slice data using a phase-field model. The proposed method is based on a modified Allen–Cahn (AC) equation with a fidelity term. The algorithm automatically generates the necessary slices between the given slices by solving the governing equation. To reconstruct a 3D volume, we first set a source slice and target slice. Next, we set the source slice as the initial condition and the target slice as the fidelity function. Finally, we retain the numerical solutions during an evolution as intermediate slices between the source and target slices. There are two criteria for choosing the intermediate slice: One is based on the area of the symmetric difference between the phase-field solution and the target and the other is based on the change of the phase-field solution relative to the area of the target. We use the weighted average of the two criteria. To validate the efficiency and accuracy of the proposed numerical algorithm, several computational experiments are conducted. Computational test results confirm the superior performance of the proposed algorithm.

Slice imputation: Multiple intermediate slices interpolation for anisotropic 3D medical image segmentation

We introduce a novel frame-interpolation-based method for slice imputation to improve segmentation accuracy for anisotropic 3D medical images, in which the number of slices and their corresponding segmentation labels can be increased between two consecutive slices in anisotropic 3D medical volumes. Unlike previous inter-slice imputation methods, which only focus on the smoothness in the axial direction, this study aims to improve the smoothness of the interpolated 3D medical volumes in all three directions: axial, sagittal, and coronal. The proposed multitask inter-slice imputation method, in particular, incorporates a smoothness loss function to evaluate the smoothness of the interpolated 3D medical volumes in the through-plane direction (sagittal and coronal). It not only improves the resolution of the interpolated 3D medical volumes in the through-plane direction but also transforms them into isotropic representations, which leads to better segmentation performances. Experiments on whole tumor segmentation in the brain, liver tumor segmentation, and prostate segmentation indicate that our method outperforms the competing slice imputation methods on both computed tomography (1\\% Dice improvement for CT liver tumor segmentation) and magnetic resonance images volumes (over 2\\% Dice improvement for MRI prostate segmentation) in most cases.

Autoencoding low-resolution MRI for semantically smooth interpolation of anisotropic MRI.

High-resolution medical images are beneficial for analysis but their acquisition may not always be feasible. Alternatively, high-resolution images can be created from low-resolution acquisitions using conventional upsampling methods, but such methods cannot exploit high-level contextual information contained in the images. Recently, better performing deep-learning based super-resolution methods have been introduced. However, these methods are limited by their supervised character, i.e. they require high-resolution examples for training. Instead, we propose an unsupervised deep learning semantic interpolation approach that synthesizes new intermediate slices from encoded low-resolution examples. To achieve semantically smooth interpolation in through-plane direction, the method exploits the latent space generated by autoencoders. To generate new intermediate slices, latent space encodings of two spatially adjacent slices are combined using their convex combination. Subsequently, the combined encoding is decoded to an intermediate slice. To constrain the model, a notion of semantic similarity is defined for a given dataset. For this, a new loss is introduced that exploits the spatial relationship between slices of the same volume. During training, an existing in-between slice is generated using a convex combination of its neighboring slice encodings. The method was trained and evaluated using publicly available cardiac cine, neonatal brain and adult brain MRI scans. In all evaluations, the new method produces significantly better results in terms of Structural Similarity Index Measure and Peak Signal-to-Noise Ratio (p<0.001 using one-sided Wilcoxon signed-rank test) than a cubic B-spline interpolation approach. Given the unsupervised nature of the method, high-resolution training data is not required and hence, the method can be readily applied in clinical settings.

Region-dependent regulation of acute ethanol on γ oscillation in the rat hippocampal slices.

Ethanol use disorders are a serious medical and public health problem in the world today. Acute ethanol intoxication can lead to cognitive dysfunction such as learning and memory impairment. Gamma oscillations (γ, 30-80Hz) are synchronized rhythmic activity generated by population of neurons within local network, and closely related to learning and memory function. The hippocampus is a critical anatomic structure that supports learning and memory. On the grounds of structure and function, hippocampus can be divided into the intermediate (IH), the dorsal (DH), and ventral hippocampus (VH). The current study is the first to investigate the effects of acute ethanol on γ oscillations in these sub-regions of rat hippocampal slices. The sustained γ oscillations were induced by 200nM kainate (KA) in the CA3c of IH, DH, and VH. When KA-induced γ oscillation reached the steady state, ethanol (50mM or 100mM) was applied and the effects of ethanol on γ oscillation power was measured in the slices sequentially sectioned from ventral to dorsal hippocampus of adult rats. In the intermediate hippocampal slices, compared with control (KA only), ethanol (50mM) caused 36.1 ± 3.9% decrease in γ power (p < 0.05, n = 10), while ethanol (100mM) caused 55.3 ± 5.5% decrease in γ power (p < 0.001, n = 14). In the dorsal hippocampus, only ethanol (100mM) caused 18.1 ± 8.6% decrease in γ power (p < 0.05, n = 12). However, in the ventral hippocampus, neither 50mM nor 100mM ethanol affected γ oscillation. Our results demonstrate that ethanol may produce the differential suppression of γ oscillations in a dose-dependent manner in different sub-regions of hippocampus, suggesting that the modulation of ethanol on hippocampal γ oscillation is region-dependent.

A smooth three-dimensional reconstruction of human head from minimally selected computed tomography slices

Three-dimensional reconstruction has been deeply investigated by researchers all over the world. This is a comprehensive effort to find an effective interpolation technique which can provide an accurate and enhanced three-dimensional reconstruction of human head from a select set of computed tomography slice data. Based on structural similarity measurement, a set of slices is selected and segmented using phase field segmentation. Keeping these segmented slices as base, the intermediate slices are re-created using linear and modified curvature registration based interpolation and the results are compared. To further enhance the result and provide a better reconstruction, we apply a refinement process using modified Cahn-Hilliard equation to the interpolated slices. The results are validated both quantitatively and qualitatively. Results show that modified curvature registration based interpolation with our proposed refinement outperforms linear interpolation with refinement providing a simultaneous improvement in sensitivity (95.95%) and specificity (95.94%) with an accuracy of more than 96% and minimal mean square error.

A hierarchical stochastic modelling approach for reconstructing lung tumour geometry from 2D CT images

ABSTRACTLung cancer is one of the deadliest cancers in both men and women. Nowadays, several methods are used to cure this cancer including surgery and radiotherapy. These methods require prior knowledge about the shape of tumours. This type of knowledge may also help physicians to determine the cancer type. In this paper we propose a novel approach for 3D reconstruction of tumour geometry from a sequence of 2D images. The proposed approach consists of two phases: tumour segmentation from computed tomography (CT) images and 3D shape reconstruction. Segmentation is conducted using snake optimisation and Gustafson–Kessel clustering. For 3D reconstruction, first, we propose a new approach to interpolate some intermediate slices between original slices. Then, the well-known marching cubes algorithm is used for surface reconstruction. Eventually, we smoothen the surface using an explicit fairing algorithm. Experiments show that our new approach can highly improve the quality and the accuracy of the reconstructed tumour shape.

Resolution enhancement of lung 4D‐CT via group‐sparsity

4D-CT typically delivers more accurate information about anatomical structures in the lung, over 3D-CT, due to its ability to capture visual information of the lung motion across different respiratory phases. This helps to better determine the dose during radiation therapy for lung cancer. However, a critical concern with 4D-CT that substantially compromises this advantage is the low superior-inferior resolution due to less number of acquired slices, in order to control the CT radiation dose. To address this limitation, the authors propose an approach to reconstruct missing intermediate slices, so as to improve the superior-inferior resolution. In this method the authors exploit the observation that sampling information across respiratory phases in 4D-CT can be complimentary due to lung motion. The authors' approach uses this locally complimentary information across phases in a patch-based sparse-representation framework. Moreover, unlike some recent approaches that treat local patches independently, the authors' approach employs the group-sparsity framework that imposes neighborhood and similarity constraints between patches. This helps in mitigating the trade-off between noise robustness and structure preservation, which is an important consideration in resolution enhancement. The authors discuss the regularizing ability of group-sparsity, which helps in reducing the effect of noise and enables better structural localization and enhancement. The authors perform extensive experiments on the publicly available DIR-Lab Lung 4D-CT dataset [R. Castillo, E. Castillo, R. Guerra, V. Johnson, T. McPhail, A. Garg, and T. Guerrero, "A framework for evaluation of deformable image registration spatial accuracy using large landmark point sets," Phys. Med. Biol. 54, 1849-1870 (2009)]. First, the authors carry out empirical parametric analysis of some important parameters in their approach. The authors then demonstrate, qualitatively as well as quantitatively, the ability of their approach to achieve more accurate and better localized results over bicubic interpolation as well as a related state-of-the-art approach. The authors also show results on some datasets with tumor, to further emphasize the clinical importance of their method. The authors have proposed to improve the superior-inferior resolution of 4D-CT by estimating intermediate slices. The authors' approach exploits neighboring constraints in the group-sparsity framework, toward the goal of achieving better localization and noise robustness. The authors' results are encouraging, and positively demonstrate the role of group-sparsity for 4D-CT resolution enhancement.

Resolution enhancement of lung 4D‐CT data using multiscale interphase iterative nonlocal means

Four-dimensional computer tomography (4D-CT) has been widely used in lung cancer radiotherapy due to its capability in providing important tumor motion information. However, the prolonged scanning duration required by 4D-CT causes considerable increase in radiation dose. To minimize the radiation-related health risk, radiation dose is often reduced at the expense of interslice spatial resolution. However, inadequate resolution in 4D-CT causes artifacts and increases uncertainty in tumor localization, which eventually results in extra damages of healthy tissues during radiotherapy. In this paper, the authors propose a novel postprocessing algorithm to enhance the resolution of lung 4D-CT data. The authors' premise is that anatomical information missing in one phase can be recovered from the complementary information embedded in other phases. The authors employ a patch-based mechanism to propagate information across phases for the reconstruction of intermediate slices in the longitudinal direction, where resolution is normally the lowest. Specifically, the structurally matching and spatially nearby patches are combined for reconstruction of each patch. For greater sensitivity to anatomical details, the authors employ a quad-tree technique to adaptively partition the image for more fine-grained refinement. The authors further devise an iterative strategy for significant enhancement of anatomical details. The authors evaluated their algorithm using a publicly available lung data that consist of 10 4D-CT cases. The authors' algorithm gives very promising results with significantly enhanced image structures and much less artifacts. Quantitative analysis shows that the authors' algorithm increases peak signal-to-noise ratio by 3-4 dB and the structural similarity index by 3%-5% when compared with the standard interpolation-based algorithms. The authors have developed a new algorithm to improve the resolution of 4D-CT. It outperforms the conventional interpolation-based approaches by producing images with the markedly improved structural clarity and greatly reduced artifacts.

Harnessing group-sparsity regularization for resolution enhancement of lung 4D-CT.

A critical concern with lung 4D-CT is the low superior-inferior resolution, due to the consideration of radiation dose. We propose a resolution enhancement approach that reconstructs missing intermediate slices by exploiting the idea that information lost in one respiratory phase can be found in others, according to the complimentary nature of inter-phase information. Our approach is based on a patch-based framework that explores the role of group-sparsity involving groups of similar neighbouring patches. We discuss the regularizing role of group-sparsity, which helps in reducing the effect of noise and enables better enhancement of anatomical structures. Our results positively demonstrate the potential of group-sparsity for 4D-CT resolution enhancement.

Hierarchical patch-based sparse representation--a new approach for resolution enhancement of 4D-CT lung data.

4D-CT plays an important role in lung cancer treatment because of its capability in providing a comprehensive characterization of respiratory motion for high-precision radiation therapy. However, due to the inherent high-dose exposure associated with CT, dense sampling along superior-inferior direction is often not practical, thus resulting in an inter-slice thickness that is much greater than in-plane voxel resolutions. As a consequence, artifacts such as lung vessel discontinuity and partial volume effects are often observed in 4D-CT images, which may mislead dose administration in radiation therapy. In this paper, we present a novel patch-based technique for resolution enhancement of 4D-CT images along the superior-inferior direction. Our working premise is that anatomical information that is missing in one particular phase can be recovered from other phases. Based on this assumption, we employ a hierarchical patch-based sparse representation mechanism to enhance the superior-inferior resolution of 4D-CT by reconstructing additional intermediate CT slices. Specifically, for each spatial location on an intermediate CT slice that we intend to reconstruct, we first agglomerate a dictionary of patches from images of all other phases in the 4D-CT. We then employ a sparse combination of patches from this dictionary, with guidance from neighboring (upper and lower) slices, to reconstruct a series of patches, which we progressively refine in a hierarchical fashion to reconstruct the final intermediate slices with significantly enhanced anatomical details. Our method was extensively evaluated using a public dataset. In all experiments, our method outperforms the conventional linear and cubic-spline interpolation methods in preserving image details and also in suppressing misleading artifacts, indicating that our proposed method can potentially be applied to better image-guided radiation therapy of lung cancer in the future.

Intensity-modulated Radiotherapy Planning from Limited Anatomical Information: Is Sim-CT Sufficient for Planning Women with Breast Cancer Receiving Intensity-modulated Radiotherapy?

To investigate intensity-modulated radiotherapy (IMRT) plans for women with carcinoma of the breast, using a small number of Sim-CT slices, thus avoiding changing the patient's position and potential problems with CT capacity. Ten CT scans of women with breast cancer were obtained for use in the study. IMRT plans based on an open tangent pair and additional top-up segment fields were created using the full CT scan, and represented the gold standard treatment plan for comparison purposes. Five-slice CT simulator scans were artificially created by omitting intermediate slices from the full CT scans. Additionally, the intermediate CT slices were recreated via interpolation of the five slices using a standard interpolation algorithm. IMRT plans were created in the same way as for the full CT scans. To allow a suitable plan comparison to be made, the beam segments and monitor units were transferred to the full CT scans, and the dose distribution calculated. The interpolated five-slice plans showed no significant difference in the volume of tissue receiving dose outside the range 95-105%, compared with the IMRT plans created using the full CT data set (1.3 +/- 2.2%, P = 0.092). In contrast, the discrete slice CT simulator plans increased by 6.3 +/- 5.4%, P = 0.0054, showing a statistically significant difference in the dose distribution produced and a clinically inferior plan. Plans created using five discrete slice CT scans were inferior to full CT-derived IMRT treatment plans, and are therefore not acceptable for IMRT. However, interpolating five CT simulator slices provides adequate anatomical information to produce comparable IMRT plans to those created by full CT scans of the patient. This allows the introduction of IMRT for this patient group without the need to change treatment position to accommodate CT scanning.

Measurement of the hybrid content of heavy quarkonia using lattice nonrelativistic QCD

Using lowest-order lattice nonrelativistic QCD to create heavy meson propagators and applying the spin-dependent interaction, ${c}_{B}(\ensuremath{-}{g/2m}_{q})\stackrel{\ensuremath{\rightarrow}}{\ensuremath{\sigma}}\ensuremath{\cdot}\stackrel{\ensuremath{\rightarrow}}{B},$ at varying intermediate time slices, we compute the off-diagonal matrix element of the Hamiltonian for the quarkonium-hybrid two-state system. Thus far, we have results for one set of quenched lattices with an interpolation in quark mass to match the bottomonium spectrum. After diagonalization of the two-state Hamiltonian, we find the ground state of the $\ensuremath{\Upsilon}$ to show a ${0.0035(1)c}_{B}^{2}$ (with ${c}_{B}^{2}\ensuremath{\sim}1.5--3.1)$ probability admixture of hybrid, $|b\overline{b}g〉.$

Generalized cones from serial sections

Slices through three-dimensional objects are a common form of data in biomedical research and practice. We report here methods for extracting high-level shape descriptors from slice data. Objects are described as generalized cones and relations between them. The cones used have straight-line spines, elliptical cross sections, and a sweeping rule whereby the lengths of the major and minor axes of the ellipses change linearly and independently. Most smooth elongated objects can be described as compositions of such cone primitives. Two algorithms were developed and tested for extracting cone descriptions from stacks of parallel slices. A local algorithm creates cones from elliptical regions on adjacent slices. A predictive algorithm hypothesises cones from regions on nonadjacent slices and then tests the hypotheses by intelligently requesting intermediate slices. Both algorithms explain complex noneliiptical regions (joints, typically) by using cones impinging upon them as keys to their shapes. Output cone sets are covers, not partitions, of the data. Topological and geometric information is available in the output description. Results of processing a “Y,” a helix, a heart phantom, and a digitized canine heart are discussed.