Low-dose computed tomography (CT) denoising is crucial for reduced radiation exposure while ensuring diagnostically acceptable image quality. Despite significant advancements driven by deep learning (DL) in recent years, existing DL-based methods, typically trained on a specific dose level and anatomical region, struggle to handle diverse noise characteristics and anatomical heterogeneity during varied scanning conditions, limiting their generalizability and robustness in clinical scenarios. In this paper, we propose FoundDiff, a foundational diffusion model for unified and generalizable LDCT denoising across various dose levels and anatomical regions. FoundDiff employs a two-stage strategy: (i) dose-anatomy perception and (ii) adaptive denoising. First, we develop a dose- and anatomy-aware contrastive language-image pre-training model (DA-CLIP) to achieve robust dose and anatomy perception by leveraging specialized contrastive learning strategies to learn continuous representations that quantify ordinal dose variations and identify salient anatomical regions. Second, we design a dose- and anatomy-aware diffusion model (DA-Diff) to perform adaptive and generalizable denoising by synergistically integrating the learned dose and anatomy embeddings from DA-CLIP into diffusion process via a novel dose and anatomy conditional block (DACB) based on Mamba. Extensive experiments on a large simulated multi-dose CT dataset spanning three anatomical regions, together with cross-dataset evaluations on Mayo-2016, CQ500, and piglet datasets, demonstrate superior denoising performance and strong generalization to unseen dose levels and anatomical regions. The codes and models are available at https://github.com/hao1635/FoundDiff.

Obtaining multiple CT scans from the same patient is required in many clinical scenarios, such as lung nodule screening and image-guided radiation therapy. Repeated scans would expose patients to higher radiation dose and increase the risk of cancer. In this study, we aim to achieve ultra-low-dose imaging for subsequent scans by collecting extremely undersampled sinogram via regional few-view scanning, and preserve image quality utilizing the preceding fullsampled scan as prior. To fully exploit prior information, we propose a two-stage framework consisting of diffusion model-based sinogram restoration and deep learning-based unrolled iterative reconstruction. Specifically, the undersampled sinogram is first restored by a conditional diffusion model with sinogram-domain prior guidance. Then, we formulate the undersampled data reconstruction problem as an optimization problem combining fidelity terms for both undersampled and restored data, along with a regularization term based on image-domain prior. Next, we propose Prior-aided Alternate Iterative NeTwork (PAINT) to solve the optimization problem. PAINT alternately updates the undersampled or restored data fidelity term, and unrolls the iterations to integrate neural network-based prior regularization. In the case of 112 mm field of view in simulated data experiments, our proposed framework achieved superior performance in terms of CT value accuracy and image details preservation. Clinical data experiments also demonstrated that our proposed framework outperformed the comparison methods in artifact reduction and structure recovery.

Cardiac CT provides comprehensive structural and functional heart imaging, but motion artifacts remain a fundamental challenge. Although modern scanners have improved temporal resolution through hardware advancements and electrocardiogram-gated scan protocols, cardiac CT is still limited to specific cardiac phases and may fail in clinical practice. Here, to overcome this challenge, we propose the MARVEL (Motion-Aware Reconstruction Via Embedded Learning of motion prior) framework for time-resolved cardiac CT imaging. The MARVEL synergistically integrates a powerful data-driven model (MP-Net) with learned Motion Prior into a robust model-based reconstruction process. This hybrid design preserves the interpretability and flexibility of model-based reconstruction methods, while benefiting from the high performance and computational efficiency of data-driven approaches. Specifically, the proposed MP-Net with specially designed mixed spatiotemporal convolutions extracts motion information from 4D pre-reconstructed images. Through an embedded learning strategy, the model learns to predict a reconstruction-oriented cardiac motion vector field (MVF). Then, a motion-aware reconstruction is implemented according to the MVF to compensate for cardiac motion. As a result, the MARVEL achieves effective reduction of motion artifacts throughout the entire heart and breaks through the conventional phase limitations of cardiac CT imaging. Furthermore, MARVEL integrates seamlessly into standard CT workflows, facilitating its potential for clinical application. Extensive evaluations, including qualitative and quantitative assessments on both simulated and clinical datasets, as well as blinded reader studies, demonstrate MARVEL's superiority over the comparison methods. Code and dynamic reconstruction demos will be available at https://github.com/ZihengD/MARVEL_CardiacCT.

Photon-counting computed tomography (PCCT) may dramatically benefit clinical practice due to its versatility such as dose reduction and material characterization. However, the limited number of photons detected in each individual energy bin can induce severe noise contamination in the reconstructed image. Fortunately, the notable low-rank prior inherent in the PCCT image can guide the reconstruction to a denoised outcome. To fully excavate and leverage the intrinsic low-rankness, we propose a novel reconstruction algorithm based on quaternion representation (QR), called low-rank quaternion reconstruction (LOQUAT). First, we organize a group of nonlocal similar patches into a quaternion matrix. Then, an adjusted weighted Schatten-p norm (AWSN) is introduced and imposed on the matrix to enforce its low-rank nature. Subsequently, we formulate an AWSN-regularized model and devise an alternating direction method of multipliers (ADMM) framework to solve it. Experiments on simulated and real-world data substantiate the superiority of the LOQUAT technique over several state-of-the-art competitors in terms of both visual inspection and quantitative metrics. Moreover, our QR-based method exhibits lower computational complexity than some popular tensor representation (TR) based counterparts. Besides, the global convergence of LOQUAT is theoretically established under a mild condition. These properties bolster the robustness and practicality of LOQUAT, facilitating its application in PCCT clinical scenarios. The source code will be available at https://github.com/linzf23/LOQUAT.

Radiation therapy is regarded as the mainstay treatment for cancer in clinic. Kilovoltage cone-beam CT (CBCT) images have been acquired for most treatment sites as the clinical routine for image-guided radiation therapy (IGRT). However, repeated CBCT scanning brings extra irradiation dose to the patients and decreases clinical efficiency. Sparse CBCT scanning is a possible solution to the problems mentioned above but at the cost of inferior image quality. To decrease the extra dose while maintaining the CBCT quality, deep learning (DL) methods are widely adopted. In this study, planning CT was used as prior information, and the corresponding strictly structure-preserved CBCT was simulated based on the attenuation information from the planning CT. We developed a hyper-resolution ultra-sparse-view CBCT reconstruction model, known as the planning CT-based strictly-structure-preserved neural network (PSSP-NET), using a generative adversarial network (GAN). This model utilized clinical CBCT projections with extremely low sampling rates for the rapid reconstruction of high-quality CBCT images, and its clinical performance was evaluated in head-and-neck cancer patients. Our experiments demonstrated enhanced performance and improved reconstruction speed.

Lung cancer screening with computed tomography (CT) scans can effectively improve the survival rate through the early detection of lung cancer, which typically identified in the form of pulmonary nodules. Multiple sequential CT images are helpful to determine nodule malignancy and play a significant role to detect lung cancers. It is crucial to develop effective lung cancer classification algorithms to achieve accurate results from multiple images without nodule location annotations, which can free radiologists from the burden of labeling nodule locations before predicting malignancy. In this study, we proposed the sequential multi-instance learning (SMILE) framework to predict high-risk lung cancer patients with multiple CT scans. SMILE included two steps. The first step was nodule instance generation. We employed the nodule detection algorithm with image category transformation to identify nodule instance locations within the entire lung images. The second step was nodule malignancy prediction. Models were supervised by patient-level annotations, without the exact locations of nodules. We embedded multi-instance learning with temporal feature extraction into a fusion framework, which effectively promoted the classification performance. SMILE was evaluated by five-fold cross-validation on a 925-patient dataset (182 malignant, 743 benign). Every patient had three CT scans, of which the interval period was about one year. Experimental results showed the potential of SMILE to free radiologists from labeling nodule locations. The source code will be available at https://github.com/wyzhao27/SMILE.

Locating the start, apex and end keyframes of moving contrast agents for keyframe counting in X-ray coronary angiography (XCA) is very important for the diagnosis and treatment of cardiovascular diseases. To locate these keyframes from the class-imbalanced and boundary-agnostic foreground vessel actions that overlap complex backgrounds, we propose long short-term spatiotemporal attention by integrating a convolutional long short-term memory (CLSTM) network into a multiscale Transformer to learn the segment- and sequence-level dependencies in the consecutive-frame-based deep features. Image-to-patch contrastive learning is further embedded between the CLSTM-based long-term spatiotemporal attention and Transformer-based short-term attention modules. The imagewise contrastive module reuses the long-term attention to contrast image-level foreground/background of XCA sequence, while patchwise contrastive projection selects the random patches of backgrounds as convolution kernels to project foreground/background frames into different latent spaces. A new XCA video dataset is collected to evaluate the proposed method. The experimental results show that the proposed method achieves a mAP (mean average precision) of 72.45% and a F-score of 0.8296, considerably outperforming the state-of-the-art methods. The source code is available at https://github.com/Binjie-Qin/STA-IPCon.

Involuntary motion of the heart remains a challenge for cardiac computed tomography (CT) imaging. Although the electrocardiogram (ECG) gating strategy is widely adopted to perform CT scans at the quasi-quiescent cardiac phase, motion-induced artifacts are still unavoidable for patients with high heart rates or irregular rhythms. Dynamic cardiac CT, which provides functional information of the heart, suffers even more severe motion artifacts. In this paper, we develop a deep learning based framework for motion artifact reduction in dynamic cardiac CT. First, we build a PAD (Pseudo All-phase clinical-Dataset) based on a whole-heart motion model and single-phase cardiac CT images. This dataset provides dynamic CT images with realistic-looking motion artifacts that help to develop data-driven approaches. Second, we formulate the problem of motion artifact reduction as a video deblurring task according to its dynamic nature. A novel TT U-Net (Temporal Transformer U-Net) is proposed to excavate the spatiotemporal features for better motion artifact reduction. The self-attention mechanism along the temporal dimension effectively encodes motion information and thus aids image recovery. Experiments show that the TT U-Net trained on the proposed PAD performs well on clinical CT scans, which substantiates the effectiveness and fine generalization ability of our method. The source code, trained models, and dynamic demo will be available at https://github.com/ivy9092111111/TT-U-Net.

Although robust PCA has been increasingly adopted to extract vessels from X-ray coronary angiography (XCA) images, challenging problems such as inefficient vessel-sparsity modelling, noisy and dynamic background artefacts, and high computational cost still remain unsolved. Therefore, we propose a novel robust PCA unrolling network with sparse feature selection for super-resolution XCA vessel imaging. Being embedded within a patch-wise spatiotemporal super-resolution framework that is built upon a pooling layer and a convolutional long short-term memory network, the proposed network can not only gradually prune complex vessel-like artefacts and noisy backgrounds in XCA during network training but also iteratively learn and select the high-level spatiotemporal semantic information of moving contrast agents flowing in the XCA-imaged vessels. The experimental results show that the proposed method significantly outperforms state-of-the-art methods, especially in the imaging of the vessel network and its distal vessels, by restoring the intensity and geometry profiles of heterogeneous vessels against complex and dynamic backgrounds. The source code is available at https://github.com/Binjie-Qin/RPCA-UNet.

Computed tomography angiography (CTA) is a powerful tool for the diagnosis of vascular diseases and its radiation dose is widely concerned. Limited scan within vessel region, deemed as region-of-interest (ROI), is a particularly apt dose reduction strategy for CTA since clinical assessment mainly depends on vessel structures. However, insufficient raw data may induce noise and artifacts in images reconstructed by conventional analytic and iterative methods. In this paper, we introduced an ultra-low-dose scan protocol for CTA, by modifying the contrast-enhanced stage with a low-mA, few-view ROI scan. Accordingly, a previous-stage-based ROI reconstruction method (PSBROI) was proposed with weighted projections and voxels and a dual-dictionary learning (DDL) strategy. Experiments showed that under the ultra-low-dose scan protocol, the proposed method performed better in artifact removal, noise suppression and structure preservation within ROI than the conventional methods. The ultra-low-dose scan with 30 mAs, 60 projection views and 34.1 mm ROI size could reduce dose to about 0.305% of normal dose.

Automated identification and localization of vertebrae in spinal computed tomography (CT) imaging is a complicated hybrid task. This task requires detecting and indexing a long sequence in a 3-D image, and both image feature extraction and sequence modeling are needed to address the problem. In this paper, the powerful fully convolutional neural network (FCN) technique performs both of these tasks simultaneously because FCNs directly encode and decode the spatial interdependence of different components in images. The key module of our proposed framework is a 3-D FCN trained in an end-to-end manner at the spine level to capture the long-range contextual information in CT volumes. The large increase in the calculation due to the full-size image inputs is alleviated by the scale-down of the inputs and the use of an auxiliary FCN to compensate for the loss of details. The composite network pipeline design enables the integration of local image details and global image patterns. Furthermore, explicit spatial and sequential constraints are imposed by the hidden Markov model (HMM) for a higher robustness and a clearer interpretation of network outputs. The proposed framework is quantitatively evaluated on the public dataset from the MICCAI 2014 Computational Challenge on Vertebrae Localization and Identification and demonstrates an identification rate (within 20 mm) of 94.67%, a mean identification rate of 87.97%, and a mean error distance of 2.56 mm on the test set, thus achieving the highest performance reported on this dataset.

Computer-aided detection (CAD) systems can assist radiologists in reducing the interpretation time and improving the detection results in computed tomographic colonography (CTC). However, existing false positives (FPs) impair the advantages of CAD systems. This study aims to develop new morphological features for the FP reduction while maintaining high detection sensitivity. Volumetric feature maps are computed for each polyp candidate by using three-dimensional (3-D) geodesic distance transformation, circular transformation (CcT), and quantized convergence index (QCI) filters. Then, new morphological features are developed based on the curvature, fractal dimension, and volumetric feature maps. To the best of our knowledge, we are also the first to develop 3-D CcT and QCI filters specifically for colonic polyps. The new morphological features were evaluated to reduce the FPs by using 456 oral contrast-enhanced CT scans from 228 patients with 130 polyps ≥5 mm. For comparison, the well-defined features from our previous work were used to generate a baseline reference. The additional use of the new morphological features reduced the FP rate from 4.2 to 2.0 FPs per scan (i.e., 52.4% FP reduction percentage) at 96.2% by-polyp sensitivity and from 4.5 to 2.1 FPs per scan (i.e., 53.3% FP reduction percentage) at 93.9% per-scan sensitivity for polyps ≥5 mm. Experimental results indicate that the new morphological features can effectively reduce the FP rate without sacrificing detection sensitivity. We believe that the newly developed morphological features would advance the CAD systems to assist radiologists in interpreting CTC images.

Compared to external device based approaches, a data-driven gating technique in PET imaging is advantageous as it does not require additional hardware or procedure. Currently, data-driven cardiac gating is less studied than respiratory gating. The aim of this paper is to develop a robust data-driven cardiac gating approach for clinical application. First, the central location of the heart is obtained from the corresponding CT image. A cylinder-shaped volume of interest (VOI) centered at the central location of the heart is used to confine cardiac signal calculation. The cardiac signal modeling the expansion/contraction of the heart is calculated using the second order moment of the tracer distribution in the VOI in the projection domain. The signal-to-noise ratio (SNR) of the cardiac motion signal is defined as the energy of the cardiac frequency components over the energy of other non-cardiac frequencies. The optimal cardiac signal with maximal SNR is obtained through an iterative optimization of signal extraction parameters. To validate our method, simulations of different scan parameters including tracer uptake and noise level were generated from the 4D XCAT phantom. Quantitative evaluation was achieved by comparing the extracted signal with the truth in the simulation study. Our method was also applied to 19 patients with high myocardium uptake and compared with the conventional center-of-mass based data-driven method. The simulation study suggests that two major limiting factors in the performance of our method are the myocardium/body uptake ratio and the count rate of the whole field of view. High accuracy of the detected signal was observed with myocardium/body uptake ratio > 7 and count rate > 100 counts/ms. Cardiac peak frequencies were successfully detected in all 19 patients using our method, while the conventional method obtained peaks in only 12 data sets. Our method was also visually validated by gated reconstructions. In summary, we have developed a novel dedicated data-driven cardiac gating method for PET by tracking the contraction/expansion of the heart during the scan. Quantitative evaluation using simulations and qualitative validation with clinical datasets both demonstrate that our method is a robust alternative to the device-based method with a high success rate.

Repeated CT scans are known to increase the risk of cancer; thus, it is paradoxical to use multiple follow-up CT scans to monitor the development of a lung nodule and conduct early treatment of the nodule. In the case of a solitary lung nodule, regional scanning and region of interest (ROI) reconstruction are likely to restore the internal area at the nodule. A limited-range few-view CT is proposed in this paper for lung nodule follow-ups with extremely reduced X-radiation. For a planned scanning of an ROI, where a solitary lung nodule is positioned, a limited-range few-view CT can be employed, and thus, less tissue is exposed to X-radiation per view. An ROI reconstruction method is also proposed that makes full use of the former standard lung scan. The experimental results show that the nodule size and shape are preserved. In the case of a 40-mm ROI, the number of exposed X-rays can be reduced by 99.6% for a circular scan and 99.9% for a 3-D scan.

Spectral computed tomography (CT) has attracted considerable attention because of its energy-resolving capability in identifying and discriminating materials. The use of a narrow energy bin can improve energy resolution. However, a narrow energy bin has high noise ratio, which degrades the imaging quality of spectral CT. To address this problem, this study exploits the structure correlations of images in the energy domain and proposed two types of united iterative reconstruction (UIR) algorithms. One type uses the well-reconstructed broad-spectrum image, with all available photons, as a constraint, whereas the other type uses a pseudo narrow-energy image, which is estimated with the use of our proposed structure-coupling (SC) method, as a constraint. The SC method utilizes local structures to connect images that are reconstructed with broad-spectrum and narrow-energy CT datasets. Given a broad-spectrum image, the SC method can accurately estimate its corresponding narrow-energy image. Results show that UIR algorithms significantly outperform conventional iterative reconstruction algorithms for narrow-energy image reconstruction in spectral CT. Among the UIR algorithms, SC-UIR yields the best results.