1 June 2001 in Materials Science and Engineering: B
19 March 2024 in IEEE Transactions on Instrumentation and Measurement
The guideline of "as low as reasonably achievable" (ALARA) for radiation dose has attracted attention to sparse-view spectral computed tomography (CT) imaging. Any missing scanning view in any energy will reduce the quality of image reconstruction and material decomposition. Recently, a series of achievements have been made in optimizing spectral CT imaging based on traditional iterative models or deep learning methods. However, these works are independent or simply coupled, often neglecting the dependency relationships in spectral CT imaging. Moreover, interpretability and generalization ability are still the challenges facing the existing methods. Therefore, we combine the advantages of traditional models and deep learning methods to propose an interpretable Hybrid-domain Integrative Transformer Iterative Network (HITI-Net) model for synchronously optimizing spectral CT reconstruction and material decomposition of sparse-view measurements. The model fully utilizes prior information in the fields of projection, image, and material, while introducing the vision transformer to learn global and long-range image information interactions. Then, the interpretable objective function is solved by the alternating directions method of multipliers (ADMM), and further optimized the network structure according to the derived expression. In addition, hybrid-domain sparse regularization terms and balance parameters are designed to adaptively learn during the training phase to enhance the generalization ability of the HITI-Net. In simulation and real preclinical experiments, the imaging results of HITI-Net are superior to other comparison methods, with more small features and sharp edges.
25 October 2024 in IEEE Transactions on Instrumentation and Measurement
Low-dose and photon-counting computed tomography (CT) denoising is a challenging task in medical imaging that has attracted significant attention. The supervised deep denoising method has significant effects, but requires a large amount of noisy-clean image pairs, which are often not available in practice. In this paper, we develop a novel self-supervised method via similarity-based visual blind-spot scheme (SVB) for real CT image denoising. Firstly, a suitable paired similar images is constructed by analyzing the characteristics between similar images at different intervals, which relaxes the condition of noise independence between them. Secondly, we introduce the blind-spot scheme into similarity-based self-supervised methods, associating blind-spot denoising with original noisy image denoising to eliminate independent noise and avoid the loss of valuable pixel information. Thirdly, to improve the generalization of the network, we design a denoising network architecture that can estimate unknown noise levels. Experimentally, the SVB method is superior to supervised methods and other self-supervised methods to a large extent, and has excellent denoising performance while ensuring the restoration of intrinsic features.
27 July 2024 in Computers in Biology and Medicine
17 November 2025 in Physics in Medicine & Biology
Objective. Spectral computed tomography (CT) is a critical tool in clinical practice, offering capabilities in multi-energy spectrum imaging and material identification. The limited-angle (LA) scanning strategy has attracted attention for its advantages in fast data acquisition and reduced radiation exposure, aligning with the as low as reasonably achievable principle. However, most deep learning-based methods require separate models for each LA setting, which limits their flexibility in adapting to new conditions. In this study, we developed a novel visual-language model (VLMs)-assisted spectral CT reconstruction (VLSR) method to address LA artifacts and enable multi-setting adaptation within a single model. Approach. The VLSR method integrates the image-text perception ability of VLMs and the image generation potential of diffusion models. Prompt engineering is introduced to better represent LA artifact characteristics, further improving artifact accuracy. Additionally, a collaborative sampling framework combining data consistency, low-rank regularization, and image-domain diffusion models is developed to produce high-quality and consistent spectral CT reconstructions. Main results. The performance of VLSR is superior to other comparison methods. Under the scanning angles of 90° and 60° for simulated data, the VLSR method improves peak signal noise ratio by at least 0.41 dB and 1.13 dB compared with other methods. Significance. VLSR method can reconstruct high-quality spectral CT images under diverse LA configurations, allowing faster and more flexible scans with dose reductions.