PWD: Prior-aware Wavelet Diffusion for Efficient Dental Limited-angle CT Reconstruction.

Computed tomography (CT), as a noninvasive method first invented in the early 1970s, has become a crucial technique in medical imaging with a variety of applications including detections of cancer, cardiovascular, neurological disorders, and so on. However, conventional CT scans are inherited with the issue of a high radiation dose, which may raise health concerns against radiologists and patients. Thus, CT radiation dose is intentionally decreased in exchange for human well-being. However, a lower radiation dose causes the problem of producing a CT image with degraded quality compared with a normal dose. Furthermore, quick and small-scale CT scans are often required when young patients or dynamic organs are being inspected. Therefore, CT scans may fail to capture full-angular coverage or acquire a sufficient number of scans for the patients, thereby having limited-angle and sparse-view of CT images, which makes it hard for them to reflect the accurate features of the target object. To fix these issues while not raising the time and amount of dose of CT scans, low-dose computed tomography (LDCT), limited-angle computed tomography (LACT), and sparse-view computed tomography have emerged as promising solutions for CT image reconstruction. However, traditional approaches in implementing LDCT, LACT, and sparse-view CT may struggle with insufficient data or secondary artifacts. Therefore, it is a foreseeable trend for LDCT, LACT, and sparse-view CT to combine with deep learning so that less data is required to reconstruct CT images with fewer secondary artifacts and improved qualities. For instance, DDPNet model with global residual learning + UNet-structured network and DuDoUFNet with blocked residual learning + UNet-structured network for LDCT; DIOR with residual block + UNet-structured network and DOLCE with denoising diffusion probabilistic model (DDPM) for LACT; PINER with black-box deep model + physical-consistency optimization and DreamNet with residual block + UNet-structured network for sparse-view CT. This paper aims to introduce state-of-the-art deep models in LDCT, LACT, and sparse-view CT based on workflow and data analysis.

Limited-angle computed tomography (LACT) improves temporal resolution and reduces radiation dose, but suffers from severe artifacts due to missing projections. Clinical workflows record abundant patient- and acquisition-level metadata, yet such information remains underutilized in image reconstruction. To tackle the ill-posed LACT inverse problem, we propose a metadata-guided two-stage diffusion framework that leverages structured clinical contexts as semantic priors for robust reconstruction. In Stage-I, we learn a metadata-to-anatomy generative prior by conditioning a transformer-based diffusion model on clinical metadata (acquisition parameters, patient demographics, and diagnostic impressions), and sampling a coarse anatomical estimate from Gaussian noise. In Stage-II, a second conditional diffusion model performs coarse-to-fine refinement, using the Stage-I estimate as an image prior while re-injecting the same metadata to recover full-resolution anatomy. To preserve anatomical fidelity and suppress hallucinations, projection-domain data consistency is enforced periodically after denoising update via an ADMM-based solver. Experiments on the public multimodal CTRATE dataset demonstrate that the proposed framework outperforms iterative, CNN-based, and diffusion-based baselines, with the greatest gains under severe truncation, e.g., up to 5.23%/11.21% higher SSIM/PSNR than the strongest metadata-free diffusion competitor at 90°. On real clinical cardiac CT, it yields coronary artery calcium scores closer to full-view references, indicating improved clinical utility. Furthermore, the proposed method is generalized to out-of-distribution angular ranges and projection geometries, and ablation results confirm complementary contributions from different metadata types under limited-angle conditions. Our results highlight clinical metadata as actionable semantic priors to synergize with physics-informed constraints to improve both reconstruction fidelity and clinical quantification in LACT.

Limited-angle reconstruction is a very important but challenging problem in the field of computed tomography (CT) and it has been extensively studied for many years. However, some difficulties still remain. Based on the theory of visible and invisible boundary developed by Quinto et.al, we propose a reconstruction model for limited-angle CT, which encodes the visible edges as priors to recover the invisible ones. The new model introduces an anisotropic ep (0

In some computed tomography (CT) applications, limited-angle projections are used for image reconstruction, and traditional reconstruction methods, such as filtered back-projection (FBP) or simultaneous algebraic reconstruction technique (SART), cannot reconstruct high-quality CT images without prior knowledge assistance. For limited-angle CT reconstruction, total variation minimization (TVM) method is not conductive to recovering image structures. Image reconstruction methods based on anisotropic total variation (ATV) and reweighted anisotropic total variation (RwATV) can significantly reduce the shading artifacts using prior knowledge of the scanning angular range and image sparsity. However, using the knowledge of image sparsity does not further improve image quality in some applications. In this paper, we propose a new reconstruction method based on anisotropic relative total variation (ARTV) for limited-angle CT reconstruction. In ARTV, the windowed inherent variation (WIV) indicates the strength of structure information and WIV values are adaptively determined by local structure information. In limited-angle CT, particular scanning angular range urges us into exerting different strengths on different directions of ARTV. Experiments on FORBILD HEAD phantom, a thoracic image and real projections of a walnut are performed to test the new method. Experimental results show that the destroyed structures are recovered to some extent, and we acquire higher image quality compared to some existing limited-angle CT reconstruction methods.

Both conventional and limited-angle computed-tomography (CT) are ill-posed problems, but where conventional CT has a small null space with few dimensions, limited-angle CT has a much larger null space and is severely ill-posed. The nature of the ill-posedness of limited-angle CT dictates that any limited-angle reconstruction method must interpolate from partial data in the Fourier domain. Large aluminum-core sandwich structures, such as those of aircraft control surfaces, simultaneously prevent full range data acquisition for CT and have face sheets that lie primarily in the limited-angle Radon transform null space. Consequently, interpolation of missing data is not possible and limited-angle CT for large sandwich structures is not feasible. This paper presents a novel method for limited-angle CT that addresses this problem. The method uses constraint-based data fusion to reduce the null space of the problem in a manner consistent with the specimen. Measurements of spatial support and face sheet thickness provide data for the additional constraints. By using the method, accurate reconstruction of sandwich structures becomes possible, where it would otherwise be impossible. The method of convex projections solves the inverse problem posed by the fusion system well because it is fast (for this problem) and easily incorporates the fusion constraints. Experimental results based on synthetic data show the improvement in reconstruction accuracy obtained by the fusion method. Real data results show the successful application of the method to a plexiglass phantom.

Objective. Due to the incomplete projection data, the image reconstructed by limited-angle computed tomography (CT) usually suffers from significant artifacts, by which the structural details are heavily blurred. In this study, we aim to develop a novel approach to improve the limited-angle CT reconstruction performance, especially for the narrow scanning angular range. Approach. A deep learning based iterative framework for limited-angle tomography is proposed, which is named multi-scale dilated dense reconstruction network (MSDDRNet). The MSDDRNet utilizes a multi-scale dilated dense convolution neural network (MSDD-CNN) with conventional reconstruction algorithm for predicting image from incomplete projection data. The MSDD-CNN enhances the image features in the network by merging the DenseNet-Like structure, which serves to restore invisible singularities and reduce artifacts, as well as introducing constraints on the projection domain data into the iterative process to achieve better image detail recovery. Additionally, to improve the training speed of the network, we use a strategy of pre-training and model migration. Main results. Numerical experiments demonstrate that the proposed MSDDRNet performs well in terms of artifact correction, noise reduction and structure recovery compared to existing methods with limited scan angles, and we also extend the proposed method to more general scanning condition and other application such as dental CT data. Significance. The proposed method is a general framework, which can be applied to other CT problems, such as low dose CT, sparse-data CT and spectral CT.

Combined spatial temporal based In-loop filter for scalable extension of HEVC

According to the designed characteristics and demands of aeronautic image compression system, lifting scheme wavelet and SPIHT algorithm was selected as the key part of software implementation, which was introduced with details. In order to improve execution efficiency, border processing was simplified reasonably and SPIHT (Set Partitioning in Hierarchical Trees) algorithm was also modified partly. The results showed that the selected scheme has a 0.4dB improvement in PSNR(peak-peak-ratio) compared with classical Shaprio's scheme. To improve the operating speed, the hardware system was then designed based on DSP and many optimization measures were then applied successfully. Practical test showed that the system can meet the real-time demand with good quality of reconstruct image, which has been used in an aeronautic image compression system practically.

There have been increasing market demands to develop more effective schemes for applying MPEG video coding standard to meet the challenges from the emerging applications. We present a real-time multi-pass MPEG video coding sys tem for constant bit rate (CBR) applications, such as Video-on-Demand (VoD) over ADSL. This scheme employs two MPEG encoders. Both encoders operate at the same CBR. The second encoder has a buffer to delay the input by an amount of time relative to the first encoder to create a look-ahead window. In encoding, the first encoder collects statistical and rate-quality characteristic information. An on-line processor then uses the collected information to derive the best coding strategy for the second encoder to encode the incoming frames in the look-ahead window. The second encoder uses the encoding parameters from the processor as the coding guide to execute the coding strategy and produce the compressed bitstream. Results on an IBM MPEG-2 encoder testing system with a 15 frame look-ahead window have shown that the scheme can obtain 0.6 to 1.6dB improvement in PSNR over the conventional single encoder on MPEG test video sources. Moreover, measured by the Tektronix Inc.'s Picture Quality Analysis (PQA) System, this scheme achieved 0.87 ~ 1.35 reduction in PQA index. The visual quality improvment is also very evident.

Compressive imaging systems with spatial-temporal encoding can be used to capture and reconstruct fast-moving objects. The imaging quality highly depends on the choice of encoding masks and reconstruction methods. In this paper, we present a new network architecture to jointly design the encoding masks and the reconstruction method for compressive high-frame-rate imaging. Unlike previous works, the proposed method takes full advantage of denoising prior to provide a promising frame reconstruction. The network is also flexible enough to optimize full-resolution masks and efficient at reconstructing frames. To this end, we develop a new dense network architecture that embeds Anderson acceleration, known from numerical optimization, directly into the neural network architecture. Our experiments show the optimized masks and the dense accelerated network respectively achieve 1.5 dB and 1 dB improvements in PSNR without adding training parameters. The proposed method outperforms other state-of-the-art methods both in simulations and on real hardware. In addition, we set up a coded two-bucket camera for compressive high-frame-rate imaging, which is robust to imaging noise and provides promising results when recovering nearly 1,000 frames per second.

A practical impossibility of prediction of signs of DCT coefficients is generally accepted. Therefore each coded sign of DCT coefficients occupies usually 1 bit of memory in compressed data. At the same time data of all coded signs of DCT coefficients occupy about 20-25% of a compressed image. In this work we propose an effective approach to predict signs of DCT coefficients in block based image compression. For that, values of pixels of already coded/decoded neighbor blocks of the image are used. The approach consist two stages. At first, values of pixels of a row and a column which both are the nearest to already coded neighbor blocks are predicted by a context-based adaptive predictor. At second stage, these row and column are used for prediction of the signs of the DCT coefficients. Depending on complexity of an image proposed method allows to compress signs of DCT coefficients to 60-85% from their original size. It corresponds to increase of compression ratio of the entire image by 3-9% (or up to 0.5 dB improvement in PSNR).

ABSTRACTUnderwater images often suffer from visual degradation, affecting downstream tasks. While recent underwater image enhancement (UIE) techniques have made some advances benefiting from deep neural networks, challenges remain in restoring fine details and achieving computational efficiency. Inspired by the success of diffusion models in image generation, we propose the Underwater Laplacian‐Guided Diffusion Model (ULDM), which enhances image features layer‐by‐layer based on the hierarchical structure of the Laplacian pyramid transform to achieve both high‐quality and efficient UIE. The Laplacian pyramid decomposes the degraded image into high‐ and low‐frequency components, enabling the model to denoise the low‐frequency spectrum and address global image degradation, thereby reducing computational overhead. To efficiently enhance high‐frequency details, we introduce the Hierarchical Attention Weighted Module (HAWM) that leverages the strong pixel correlations in high‐frequency sub‐images at different levels, adjusting them layer‐by‐layer to better capture fine details. These high‐frequency sub‐images exhibit strong pixel correlation and consistent texture features across different layers, and their hierarchical pattern ensures effective detail restoration. Extensive experiments demonstrate that ULDM outperforms state‐of‐the‐art methods in both quantitative and qualitative evaluations.

Although terahertz (THz) medical imaging is advantageous owing to its non-ionizing nature and molecular sensitivity, signal attenuation resulting from high water content and diverse noise sources necessitates effective denoising. Deep learning models demonstrate notable advantages in medical image denoising. However, supervised models rely on paired clean-noisy data, which deviates from clinical reality, while self-supervised models eliminate the need for paired data but suffer from a discrepancy between learned and real-world image priors, leading to performance degradation. To overcome these limitations, this paper enhances DnCNN by integrating the Squeeze-and-Excitation (SE) block, which adaptively recalibrates channel-wise feature responses to emphasize informative regions and improve contrast between structures, and employs the Noisy-As-Clean (NAC) strategy for self-supervised training, which effectively narrows the disparity between training and test conditions by adapting the model directly to the characteristics of the test image itself. Experimental results on our own THz medical image dataset, which comprises 320 PNG format images of 369×369 pixels generated from 3D-printed tooth models using THz imaging, demonstrate significant quantitative improvements in PSNR and SSIM compared with other renowned denoising algorithms. These results confirm that the integrated SE block effectively enhances edge and texture preservation, while the NAC strategy substantially reduces dependency on paired data and boosts overall denoising performance.

Denoising computed tomography (CT) medical images is crucial in preserving information and restoring images contaminated with noise. Standard filters have extensively been used for noise removal and fine details’ preservation. During the transmission of medical images, noise degrades the visibility of anatomical structures and subtle abnormalities, making it difficult for radiologists to accurately diagnose and interpret medical conditions. In recent studies, an optimum denoising filter using the wavelet threshold and deep-CNN was used to eliminate Gaussian noise in CT images using the image quality index (IQI) and peak signal-to-noise ratio (PSNR). Although the results were better than those with traditional techniques, the performance resulted in a loss of clarity and fine details’ preservation that rendered the CT images unsuitable. To address these challenges, this paper focuses on eliminating noise in CT scan images corrupted with additive Gaussian blur noise (AGBN) using an ensemble approach that integrates anisotropic Gaussian filter (AGF) and wavelet transform with a deep learning denoising convolutional neural network (DnCNN). First, the noisy image is denoised by AGF and Haar wavelet transform as preprocessing operations to eliminate AGBN. The DnCNN is then combined with AGF and wavelet for post-processing operation to eliminate the rest of the noises. Specifically, we used AGF due to its adaptability to edge orientation and directional information, which prevents blurring along edges for non-uniform noise distribution. Denoised images are evaluated using PSNR, mean squared error (MSE), and the structural similarity index measure (SSIM). Results revealed that the average PSNR value of the proposed ensemble approach is 28.28, and the average computational time is 0.01666 s. The implication is that the MSE between the original and reconstructed images is very low, implying that the image is restored correctly. Since the SSIM values are between 0 and 1.0, 1.0 perfectly matches the reconstructed image with the original image. In addition, the SSIM values at 1.0 or near 1.0 implicitly reveal a remarkable structural similarity between the denoised CT image and the original image. Compared to other techniques, the proposed ensemble approach has demonstrated exceptional performance in maintaining the quality of the image and fine details’ preservation.

Local differential privacy algorithms combined with deep generative models can enhance secure medical image sharing among researchers in the public domain without central administrators; however, these images were limited to the generation of low-resolution images, which are very insufficient for diagnosis by medical doctors. To enhance the performance of deep generative models so that they can generate high-resolution medical images, we propose a large-scale diffusion model that can, for the first time, unconditionally generate high-resolution (256×256×256) volumetric medical images (head magnetic resonance images). This diffusion model has 19 billion parameters, but to make it easy to train it, we temporally divided the model into 200 submodels, each of which has 95 million parameters. Moreover, on the basis of this new diffusion model, we propose another formulation of image anonymization with which the processed images can satisfy provable Gaussian local differential privacy and with which we can generate images semantically different from the original image but belonging to the same class. We believe that the formulation of this new diffusion model and the implementation of local differential privacy algorithms combined with the diffusion models can contribute to the secure sharing of practical images upstream of data processing.