Senrong You;Bin Yuan;Zhihan Lyu;Charles K. Chui;C. L. Philip Chen;Baiying Lei;Shuqiang Wang
{"title":"Generative AI Enables Synthesizing Cross-Modality Brain Image via Multi-Level-Latent Representation Learning","authors":"Senrong You;Bin Yuan;Zhihan Lyu;Charles K. Chui;C. L. Philip Chen;Baiying Lei;Shuqiang Wang","doi":"10.1109/TCI.2024.3434724","DOIUrl":"10.1109/TCI.2024.3434724","url":null,"abstract":"Multiple brain imaging modalities can provide complementary pathologic information for clinical diagnosis. However, it is huge challenge to acquire enough modalities in clinical practice. In this work, a cross-modality reconstruction model, called fine-grain aware generative adversarial network (FA-GAN), is proposed to reconstruct the target modality images of brain from the 2D source modality images with a dual-stages manner. The FA-GAN is able to mine the multi-level shared latent representations from the source modality images and then reconstruct the target modality image from coarse to fine progressively. Specifically, in the coarse stage, the Multi-Grain Extractor firstly extracts and disentangles the shared latent features from the source modality images, and synthesizes the coarse target modality images with a pyramidal network. The Feature-Joint Encoder then encodes the latent features and frequency features jointly. In the fine stage, the Fine-Texture Generator is fed with the joint codes to fine tune the reconstruction of the fine-grained target modality. The wavelet transformation module is employed to extract the frequency codes and guide the Fine-Texture Generator to synthesize finer textures. Comprehensive experiments from MR to PET images on ADNI datasets demonstrate that the proposed model achieves finer structure recovery and outperforms the competing methods quantitatively and qualitatively.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1152-1164"},"PeriodicalIF":4.2,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141869570","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Paula Arguello;Jhon Lopez;Karen Sanchez;Carlos Hinojosa;Fernando Rojas-Morales;Henry Arguello
{"title":"Learning to Describe Scenes via Privacy-Aware Designed Optical Lens","authors":"Paula Arguello;Jhon Lopez;Karen Sanchez;Carlos Hinojosa;Fernando Rojas-Morales;Henry Arguello","doi":"10.1109/TCI.2024.3426975","DOIUrl":"10.1109/TCI.2024.3426975","url":null,"abstract":"Scene captioning consists of accurately describing the visual information using text, leveraging the capabilities of computer vision and natural language processing. However, current image captioning methods are trained on high-resolution images that may contain private information about individuals within the scene, such as facial attributes or sensitive data. This raises concerns about whether machines require high-resolution images and how we can protect the private information of the users. In this work, we aim to protect privacy in the scene captioning task by addressing the issue directly from the optics before image acquisition. Specifically, motivated by the emerging trend of integrating optics design with algorithms, we introduce a learned refractive lens into the camera to ensure privacy. Our optimized lens obscures sensitive visual attributes, such as faces, ethnicity, gender, and more, in the acquired image while extracting relevant features, enabling descriptions even from highly distorted images. By optimizing the refractive lens and a deep network architecture for image captioning end-to-end, we achieve description generation directly from our distorted images. We validate our approach with extensive simulations and hardware experiments. Our results show that we achieve a better trade-off between privacy and utility when compared to conventional non-privacy-preserving methods on the COCO dataset. For instance, our approach successfully conceals private information within the scene while achieving a BLEU-4 score of 27.0 on the COCO test set.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1069-1079"},"PeriodicalIF":4.2,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141869573","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Aleksei Sholokhov;Saleh Nabi;Joshua Rapp;Steven L. Brunton;J. Nathan Kutz;Petros T. Boufounos;Hassan Mansour
{"title":"Single-Pixel Imaging of Spatio-Temporal Flows Using Differentiable Latent Dynamics","authors":"Aleksei Sholokhov;Saleh Nabi;Joshua Rapp;Steven L. Brunton;J. Nathan Kutz;Petros T. Boufounos;Hassan Mansour","doi":"10.1109/TCI.2024.3434541","DOIUrl":"10.1109/TCI.2024.3434541","url":null,"abstract":"Imaging dynamic spatio-temporal flows typically requires high-speed, high-resolution sensors that may be physically or economically prohibitive. Single-pixel imaging (SPI) has emerged as a low-cost acquisition technique where light from a scene is projected through a spatial light modulator onto a single photodiode with a high temporal acquisition rate. The scene is then reconstructed from the temporal samples using computational techniques that leverage prior assumptions on the scene structure. In this paper, we propose to image spatio-temporal flows from incomplete measurements by leveraging scene priors in the form of a reduced-order model (ROM) of the dynamics learned from training data examples. By combining SPI acquisition with the ROM prior implemented as a neural ordinary differential equation, we achieve high-quality image sequence reconstruction with significantly reduced data requirements. Specifically, our approach achieves similar performance levels to leading methods despite using one to two orders of magnitude fewer samples. We demonstrate superior reconstruction at low sampling rates for simulated trajectories governed by Burgers' equation, Kolmogorov flow, and turbulent plumes emulating gas leaks.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1124-1138"},"PeriodicalIF":4.2,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141869571","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Region-Based Spectral-Spatial Mutual Induction Network for Hyperspectral Image Reconstruction","authors":"Jianan Li;Wangcai Zhao;Tingfa Xu","doi":"10.1109/TCI.2024.3430478","DOIUrl":"10.1109/TCI.2024.3430478","url":null,"abstract":"In hyperspectral compression imaging, the choice of a reconstruction algorithm is critical for achieving high-quality results. Hyperspectral Images (HSI) have strong spectral-spatial correlations within local regions, valuable for reconstruction. However, existing learning-based methods often overlook regional variations by treating the entire image as a whole. To address this, we propose a novel region-based iterative approach for HSI reconstruction. We introduce a deep unfolding method augmented with a Region-based Spectral-Spatial Mutual Induction (RSSMI) network to model regional priors. Our approach involves partitioning the image into regions during each unfolding phase. Within each region, we employ a spatial-guided spectral attention module for holistic spectral relationships and a spectral-guided spatial attention module for spatial details. By leveraging mutual induction, our method simultaneously recovers spectral and spatial information. Furthermore, we address the issue of favoring easy-to-learn regions by introducing Focal Region Loss that dynamically adjusts loss weights for regions, emphasizing those that are harder to reconstruct. Experimental results demonstrate that our method achieves competitive performance and excels in spectrum and texture reconstruction on both simulated and real HSI datasets.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1139-1151"},"PeriodicalIF":4.2,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141740335","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Super-Resolution in Low Dose X-ray CT via Focal Spot Mitigation With Generative Diffusion Networks","authors":"Carlos M. Restrepo-Galeano;Gonzalo R. Arce","doi":"10.1109/TCI.2024.3430487","DOIUrl":"10.1109/TCI.2024.3430487","url":null,"abstract":"Advancing the resolution capabilities of X-ray CT imaging, particularly in low-dose applications, is a paramount pursuit in the field. This quest for superior spatial detail is hindered by the pervasive issue of focal spot blooming, which plagues medical scanners due to the finite nature of the emittance surface in X-ray sources. Such a phenomenon introduces optical distortions in the measurements that limit the achievable resolution. In response to this challenge, we introduce a novel approach: Focal Spot Diffusion CT (FSD-CT). Unlike traditional methods that rely on limited and simplified idealizations of X-ray models, FSD-CT adopts a more complex, realistic representation of X-ray sources. FSD-CT leverages a generative diffusion-based reconstruction framework, guided by a forward imaging model for sample consistency and a frequency selection module for enhanced spectral content. FSD-CT successfully mitigates focal spot blooming without imposing a significant computational burden when compared to other diffusion-based reconstruction methods, offering a versatile solution for improving CT resolution. Computational experiments using simulations based on commercial medical scanners show FSD-CT delivers gains of up to 4 dB in fan-beam tomography compared to benchmarks such as filtered backprojection, end-to-end CNNs, and state-of-the-art diffusion models. The technique's robustness is confirmed in challenging scenarios, including sparse angle CT, off-distribution samples, and reconstructions from real projections. FSD-CT helps to overcome limitations in spatial resolution and offers a plausible solution that could be applied in clinical CT imaging after more in-depth studies are conducted.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1111-1123"},"PeriodicalIF":4.2,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141740334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yuanke Zhang;Rujuan Cao;Fan Xu;Rui Zhang;Fengjuan Jiang;Jing Meng;Fei Ma;Yanfei Guo;Jianlei Liu
{"title":"Side Information-Assisted Low-Dose CT Reconstruction","authors":"Yuanke Zhang;Rujuan Cao;Fan Xu;Rui Zhang;Fengjuan Jiang;Jing Meng;Fei Ma;Yanfei Guo;Jianlei Liu","doi":"10.1109/TCI.2024.3430469","DOIUrl":"10.1109/TCI.2024.3430469","url":null,"abstract":"CT images from individual patients or different patient populations typically share similar radiological features such as textures and structures. In model-based iterative reconstruction (MBIR) for low-dose CT (LDCT) imaging, image quality enhancement can be achieved not only by relying on the intrinsic raw data, but also by incorporating side information extracted from high-quality normal-dose CT (NDCT) exemplar images. The additional side information helps overcome the inherent limitations of raw data in low-dose scanning and offers potential improvements in LDCT image quality. This study investigates the effectiveness of side information-assisted MBIR (SI-MBIR) in enhancing the quality of LDCT images. Specifically, we propose to use the noise-free exemplar images to generate side information that aligns with the structural features of regions of interest (ROIs) in the target image. Each ROI is enhanced with a custom-designed prior subspace that is derived from similar exemplar samples and reflects its unique structural and textural characteristics. We then propose an adaptive sparse modeling approach, in particular, a weighted Laplace distribution model for the prior subspace. The weighted Laplace model is carefully tuned to match the signal-to-noise ratio (SNR) of each transform band, allowing adaptive sparse modeling on different bands. Furthermore, we propose an efficient CT reconstruction algorithm based on this adaptive sparse model. Using the alternating direction method of multipliers (ADMM) framework, an optimization method for this reconstruction algorithm has been formulated. Extensive experimental studies were conducted to validate the effectiveness of the proposed algorithm. The results show that the proposed algorithm can achieve noticeable improvements over some state-of-the-art MBIR methods in terms of noise suppression and texture preservation.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1080-1093"},"PeriodicalIF":4.2,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141740336","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jingchao Hou;Garas Gendy;Guo Chen;Liangchao Wang;Guanghui He
{"title":"DTDeMo: A Deep Learning-Based Two-Stage Image Demosaicing Model With Interpolation and Enhancement","authors":"Jingchao Hou;Garas Gendy;Guo Chen;Liangchao Wang;Guanghui He","doi":"10.1109/TCI.2024.3426360","DOIUrl":"10.1109/TCI.2024.3426360","url":null,"abstract":"Image demosaicing is one of the most ubiquitous and performance-critical image processing tasks. However, traditional demosaicing methods use fixed weights to finish the interpolation, while deep learning demosaicing restoration always breaks the image array arrangement rule, and they can't fully use the existing pixel information. To rectify these weaknesses, in this paper, we propose the convolution interpolation block (CIB) to obey the RAW data arrangement rule and the deep demosaicing residual block (DDRB) to repeatedly utilize existing pixel information for demosaicing. Based on the CIB and DDRB, we present a novel two-stage demosaicing model (DTDeMo), including differential interpolation and enhancement processes. Specifically, the interpolation process contains several CIBs and DDRBs with trainable interpolation parameters. Meanwhile, the enhancement process consists of a transformer-based block and a series of DDRBs to enhance the interpolation results. The effectiveness of CIBs, DDRBs, the proposed interpolation process, and the enhancement process is confirmed through an ablation study. A thorough comparison with several methods shows that our DTDeMo outperforms state of the art quantitatively and qualitatively.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1026-1039"},"PeriodicalIF":4.2,"publicationDate":"2024-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141588024","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mohammad Khateri;Morteza Ghahremani;Alejandra Sierra;Jussi Tohka
{"title":"No-Clean-Reference Image Super-Resolution: Application to Electron Microscopy","authors":"Mohammad Khateri;Morteza Ghahremani;Alejandra Sierra;Jussi Tohka","doi":"10.1109/TCI.2024.3426349","DOIUrl":"10.1109/TCI.2024.3426349","url":null,"abstract":"The inability to acquire clean high-resolution (HR) electron microscopy (EM) images over a large brain tissue volume hampers many neuroscience studies. To address this challenge, we propose a deep-learning-based image super-resolution (SR) approach to computationally reconstruct a clean HR 3D-EM image with a large field of view (FoV) from noisy low-resolution (LR) acquisition. Our contributions are I) investigation of training with no-clean references; II) introduction of a novel network architecture, named EMSR, for enhancing the resolution of LR EM images while reducing inherent noise. The EMSR leverages distinctive features in brain EM images–repetitive textural and geometrical patterns amidst less informative backgrounds– via multiscale edge-attention and self-attention mechanisms to emphasize edge features over the background; and, III) comparison of different training strategies including using acquired LR and HR image pairs, i.e., real pairs with no-clean references contaminated with real corruptions, pairs of synthetic LR and acquired HR, as well as acquired LR and denoised HR pairs. Experiments with nine brain datasets showed that training with real pairs can produce high-quality super-resolved results, demonstrating the feasibility of training with nonclean references. Additionally, comparable results were observed, both visually and numerically, when employing denoised and noisy references for training. Moreover, utilizing the network trained with synthetically generated LR images from HR counterparts proved effective in yielding satisfactory SR results, even in certain cases, outperforming training with real pairs. The proposed SR network was compared quantitatively and qualitatively with several established SR techniques, demonstrating either the superiority or competitiveness of the proposed method in recovering fine details while mitigating noise.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1094-1110"},"PeriodicalIF":4.2,"publicationDate":"2024-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10592622","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141588023","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Deep Learned Non-Linear Propagation Model Regularizer for Compressive Spectral Imaging","authors":"Romario Gualdrón-Hurtado;Henry Arguello;Jorge Bacca","doi":"10.1109/TCI.2024.3422900","DOIUrl":"10.1109/TCI.2024.3422900","url":null,"abstract":"Coded aperture snapshot spectral imager (CASSI), efficiently captures 3D spectral images by sensing 2D projections of the scene. While CASSI offers a substantial reduction in acquisition time, compared to traditional scanning optical systems, it requires a reconstruction post-processing step. Furthermore, to obtain high-quality reconstructions, an accurate propagation model is required. Notably, CASSI exhibits a variant spatio-spectral sensor response, making it difficult to acquire an accurate propagation model. To address these inherent limitations, this work proposes to learn a deep non-linear fully differentiable propagation model that can be used as a regularizer within an optimization-based reconstruction algorithm. The proposed approach trains the non-linear spatially-variant propagation model using paired compressed measurements and spectral images, by employing side information only in the calibration step. From the deep propagation model incorporation into a plug-and-play alternating direction method of multipliers framework, our proposed method outperforms traditional CASSI linear-based models. Extensive simulations and a testbed implementation validate the efficacy of the proposed methodology.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1016-1025"},"PeriodicalIF":4.2,"publicationDate":"2024-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141573765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"PGIUN: Physics-Guided Implicit Unrolling Network for Accelerated MRI","authors":"Jiawei Jiang;Zihan He;Yueqian Quan;Jie Wu;Jianwei Zheng","doi":"10.1109/TCI.2024.3422840","DOIUrl":"10.1109/TCI.2024.3422840","url":null,"abstract":"To cope with the challenges stemming from prolonged acquisition periods, compressed sensing MRI has emerged as a popular technique to accelerate the reconstruction of high-quality images from under-sampled k-space data. Most current solutions endeavor to solve this issue with the pursuit of certain prior properties, yet the treatments are all enforced in the original space, resulting in limited feature information. To boost the performance yet with the guarantee of high running efficiency, in this study, we propose a Physics-Guided Implicit Unrolling Network (PGIUN). Specifically, by an elaborately designed reversible network, the inputs are first mapped to a channel-lifted implicit space, which taps the potential of capturing spatial-invariant features sufficiently. Within this implicit space, we then unfold an accelerated optimization algorithm to iterate an efficient and feasible solution, in which a parallelly dual-domain update is equipped for better feature fusion. Finally, an inverse embedding transformation of the recovered high-dimensional representation is employed to achieve the desired estimation. PGIUN enjoys high interpretability benefiting from the physically induced modules, which not only facilitates an intuitive understanding of the internal working mechanism but also endows it with high generalization ability. Extensive experiments conducted across diverse datasets and varying sampling rates/patterns consistently establish the superiority of our approach over state-of-the-art methods in both visual and quantitative evaluations.","PeriodicalId":56022,"journal":{"name":"IEEE Transactions on Computational Imaging","volume":"10 ","pages":"1055-1068"},"PeriodicalIF":4.2,"publicationDate":"2024-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141549933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}