6 May 2024 in IEEE Transactions on Instrumentation and Measurement
Cone-beam computed tomography (CBCT) has been widely used in image-guided radiotherapy (IGRT). In order to avoid the possible collisions of the moving gantry with patients and devices, the limited-angle scanning protocol is often adopted for CBCT. However, the missing views of projection data will induce severe wedge artifacts to the reconstructed images, having a negative influence on the following therapeutic procedures. Compared to limited-angle CBCT images, the planning CT images (pCT) acquired earlier for the same patient are artifact-free, which has the potential to improve the quality of limited-angle CBCT. In this context, this article proposes the prior image-constrained adaptive step size iterative reconstruction (PICAS) method. PICAS builds on prior image-constrained compressed sensing (PICCS) but is improved after taking into account the characteristics of IGRT. In PICAS, the high-quality pCT images are regarded as prior images to enhance the performance in artifact removal. To address the mismatch between pCT images and CBCT images, the pCT images are reprojected and reconstructed using the same imaging geometry as CBCT. Then, a transformation matrix between the pseudo-limited-angle pCT images and limited-angle images is obtained and it is further applied to the high-quality pCT images to generate the artifact-free prior images. In addition, an adaptive gradient descent (GD) optimization based on convex set projection and Lipschitz constant is adopted to accelerate the convergence of the algorithm. The proposed PICAS algorithm has been evaluated on the real data from different parts (head, lung, and abdomen) and under the limited-angle configurations with different scanning ranges (80°, 100°, and 120°). Qualitative and quantitative results demonstrate that PICAS has the potential to quickly reconstruct high-quality CBCT images in limited-angle scanning scenes, and thereby improve the accuracy of radiotherapy and shorten the time of surgery.
18 October 2024 in IEEE Transactions on Medical Imaging
4D cone-beam computed tomography (CBCT) plays a critical role in adaptive radiation therapy for lung cancer. However, extremely sparse sampling projection data will cause severe streak artifacts in 4D CBCT images. Existing deep learning (DL) methods heavily rely on large labeled training datasets which are difficult to obtain in practical scenarios. Restricted by this dilemma, DL models often struggle with simultaneously retaining dynamic motions, removing streak degradations, and recovering fine details. To address the above challenging problem, we introduce a Deep Prior Image Constrained Motion Compensation framework (DPI-MoCo) that decouples the 4D CBCT reconstruction into two sub-tasks including coarse image restoration and structural detail fine-tuning. In the first stage, the proposed DPI-MoCo combines the prior image guidance, generative adversarial network, and contrastive learning to globally suppress the artifacts while maintaining the respiratory movements. After that, to further enhance the local anatomical structures, the motion estimation and compensation technique is adopted. Notably, our framework is performed without the need for paired datasets, ensuring practicality in clinical cases. In the Monte Carlo simulation dataset, the DPI-MoCo achieves competitive quantitative performance compared to the state-of-the-art (SOTA) methods. Furthermore, we test DPI-MoCo in clinical lung cancer datasets, and experiments validate that DPI-MoCo not only restores small anatomical structures and lesions but also preserves motion information.
29 May 2024 in IEEE Transactions on Instrumentation and Measurement
In image-guided radiation therapy, the on-board cone-beam computed tomography (CBCT) is usually used for volumetric imaging and the limited-angle scanning protocol is often adopted to avoid the possible collisions of the moving gantry with patients and devices. However, images directly reconstructed using incomplete projection data may suffer from severe artifacts, which cannot provide precise guidance for the following therapeutic procedures. Compared with CBCT, the planning CT (pCT) acquired for the treatment plan can provide high-quality images of the same patient, showing the potential to improve the limited-angle CBCT. In this context, we propose a Multi-dimensional Joint Cascaded Network (MJCNet) which can exploit the prior information from pCT. MJCNet improves the imaging quality of limited-angle CBCT through a coarse-to-fine strategy. In the coarse restoration stage, a 2D network with two encoders that could extract and exploit the information of pCT is used to remove limited-angle artifacts slice-by-slice. In the fine-tuning stage, a 3D network with dense attention mechanism is employed to further improve the image details and remove the inter-slice artifacts. The real data from different parts of human bodies are collected to evaluate the proposed method. Experimental results demonstrate the promising performance of MJCNet in reducing wedge artifacts, restoring image structures and correcting HU numbers.
19 January 2026 in IEEE Transactions on Circuits and Systems for Video Technology
Sparse-view computed tomography (SVCT) offers the advantages of accelerated scanning and reduced X-ray radiation dose in different clinical applications. However, it faces a challenge due to incomplete data acquisition, resulting in streak artifacts in the analytically reconstructed CT images. Utilizing self-supervised learning, implicit neural representation (INR) recently has shown great promise in addressing inverse problems such as SVCT reconstruction. Nonetheless, given that the input of original INR only contains coordinate information, it is limited to represent one SVCT instance at a time, and its performance significantly declines when performing cross-instance reconstruction. In this study, we propose a novel self-supervised framework named VFMINR, which leverages generalizable representations extracted from the visual foundation models (VFMs) to tackle the cross-instance reconstruction issue of INR. Specifically, VFMINR first utilizes VFMs to effectively capture the spatial and frequency domain representations of sinograms, and then a fusion module is applied to fuse two domain features into complementary representations. This combination maximizes the utilization of local detail information from the spatial domain and the global structural information from the frequency domain. Subsequently, an adaptive cell decoding strategy is designed to map representations into variable resolution hybrid feature grids, which are integrated into the learning of the INR to enhance its generalizability for different SVCT instances. The VFMs and VFMINR are trained by using only SV sinogram data, and extensive results confirm that the proposed method can effectively handle the generalization problem of INR, while achieving superior performance in image fidelity and artifact suppression. The code is available at: https://github.com/nightastars/VFMINR-main.