Zichun Zhong

A novel markerless technique to evaluate daily lung tumor motion based on conventional cone-beam CT projection data.

PURPOSE In this study, we present a novel markerless technique, based on cone beam computed tomography (CBCT) raw projection data, to evaluate lung tumor daily motion. METHOD AND MATERIALS The markerless technique, which uses raw CBCT projection data and locates tumors directly on every projection, consists of three steps. First, the tumor contour on the planning CT is used to create digitally reconstructed radiographs (DRRs) at every projection angle. Two sets of DRRs are created: one showing only the tumor, and another with the complete anatomy without the tumor. Second, a rigid two-dimensional image registration is performed to register the DRR set without the tumor to the CBCT projections. After the registration, the projections are subtracted from the DRRs, resulting in a projection dataset containing primarily tumor. Finally, a second registration is performed between the subtracted projection and tumor-only DRR. The methodology was evaluated using a chest phantom containing a moving tumor, and retrospectively in 4 lung cancer patients treated by stereotactic body radiation therapy. Tumors detected on projection images were compared with those from three-dimensional (3D) and four-dimensional (4D) CBCT reconstruction results. RESULTS Results in both static and moving phantoms demonstrate that the accuracy is within 1 mm. The subsequent application to 22 sets of CBCT scan raw projection data of 4 lung cancer patients includes about 11,000 projections, with the detected tumor locations consistent with 3D and 4D CBCT reconstruction results. This technique reveals detailed lung tumor motion and provides additional information than conventional 4D images. CONCLUSION This technique is capable of accurately characterizing lung tumor motion on a daily basis based on a conventional CBCT scan. It provides daily verification of the tumor motion to ensure that these motions are within prior estimation and covered by the treatment planning volume.

Anisotropic surface meshing with conformal embedding

This paper introduces a parameterization-based approach for anisotropic surface meshing. Given an input surface equipped with an arbitrary Riemannian metric, this method generates a metric-adapted mesh with user-specified number of vertices. In the proposed method, the edge length of the input surface is directly adjusted according to the given Riemannian metric at first. Then the adjusted surface is conformally embedded into a parametric 2D domain and a weighted Centroidal Voronoi Tessellation and its dual Delaunay triangulation are computed on the parametric domain. Finally the generated Delaunay triangulation can be mapped from the parametric domain to the original space, and the triangulation exhibits the desired anisotropic property. We compute the high-quality remeshing results for surfaces with different types of topologies and compare our method with several state-of-the-art approaches in anisotropic surface meshing by using the standard measurement criteria.

Deep convolutional neural network with transfer learning for rectum toxicity prediction in cervical cancer radiotherapy: a feasibility study

Better understanding of the dose-toxicity relationship is critical for safe dose escalation to improve local control in late-stage cervical cancer radiotherapy. In this study, we introduced a convolutional neural network (CNN) model to analyze rectum dose distribution and predict rectum toxicity. Forty-two cervical cancer patients treated with combined external beam radiotherapy (EBRT) and brachytherapy (BT) were retrospectively collected, including twelve toxicity patients and thirty non-toxicity patients. We adopted a transfer learning strategy to overcome the limited patient data issue. A 16-layers CNN developed by the visual geometry group (VGG-16) of the University of Oxford was pre-trained on a large-scale natural image database, ImageNet, and fine-tuned with patient rectum surface dose maps (RSDMs), which were accumulated EBRT  +  BT doses on the unfolded rectum surface. We used the adaptive synthetic sampling approach and the data augmentation method to address the two challenges, data imbalance and data scarcity. The gradient-weighted class activation maps (Grad-CAM) were also generated to highlight the discriminative regions on the RSDM along with the prediction model. We compare different CNN coefficients fine-tuning strategies, and compare the predictive performance using the traditional dose volume parameters, e.g. D 0.1/1/2cc, and the texture features extracted from the RSDM. Satisfactory prediction performance was achieved with the proposed scheme, and we found that the mean Grad-CAM over the toxicity patient group has geometric consistence of distribution with the statistical analysis result, which indicates possible rectum toxicity location. The evaluation results have demonstrated the feasibility of building a CNN-based rectum dose-toxicity prediction model with transfer learning for cervical cancer radiotherapy.

Sparse Localized Decomposition of Deformation Gradients

Sparse localized decomposition is a useful technique to extract meaningful deformation components out of a training set of mesh data. However, existing methods cannot capture large rotational motion in the given mesh dataset. In this paper we present a new decomposition technique based on deformation gradients. Given a mesh dataset, the deformation gradient field is extracted, and decomposed into two groups: rotation field and stretching field, through polar decomposition. These two groups of deformation information are further processed through the sparse localized decomposition into the desired components. These sparse localized components can be linearly combined to form a meaningful deformation gradient field, and can be used to reconstruct the mesh through a least squares optimization step. Our experiments show that the proposed method addresses the rotation problem associated with traditional deformation decomposition techniques, making it suitable to handle not only stretched deformations, but also articulated motions that involve large rotations.

4D cone-beam CT reconstruction using multi-organ meshes for sliding motion modeling.

A simultaneous motion estimation and image reconstruction (SMEIR) strategy was proposed for 4D cone-beam CT (4D-CBCT) reconstruction and showed excellent results in both phantom and lung cancer patient studies. In the original SMEIR algorithm, the deformation vector field (DVF) was defined on voxel grid and estimated by enforcing a global smoothness regularization term on the motion fields. The objective of this work is to improve the computation efficiency and motion estimation accuracy of SMEIR for 4D-CBCT through developing a multi-organ meshing model. Feature-based adaptive meshes were generated to reduce the number of unknowns in the DVF estimation and accurately capture the organ shapes and motion. Additionally, the discontinuity in the motion fields between different organs during respiration was explicitly considered in the multi-organ mesh model. This will help with the accurate visualization and motion estimation of the tumor on the organ boundaries in 4D-CBCT. To further improve the computational efficiency, a GPU-based parallel implementation was designed. The performance of the proposed algorithm was evaluated on a synthetic sliding motion phantom, a 4D NCAT phantom, and four lung cancer patients. The proposed multi-organ mesh based strategy outperformed the conventional Feldkamp-Davis-Kress, iterative total variation minimization, original SMEIR and single meshing method based on both qualitative and quantitative evaluations.

Dynamic meshing for deformable image registration

Finite element method (FEM) is commonly used for deformable image registration. However, there is no existing literature studying how the superimposed mesh structure would influence the image registration process. We study this problem in this paper, and propose a dynamic meshing strategy to generate mesh structure for image registration. To construct such a dynamic mesh during image registration, three steps are performed. Firstly, a density field that measures the importance of a pixel/voxels displacement to the registration process is computed. Secondly, an efficient contraction-optimization scheme is applied to compute a discrete Centroidal Voronoi Tessellation of the density field. Thirdly, the final mesh structure is constructed by its dual triangulation, with some post-processing to preserve the image boundary. In each iteration of the deformable image registration, the mesh structure is efficiently updated with GPU-based parallel implementation. We conduct experiments of the new dynamic mesh-guided registration framework on both synthetic and real medical images, and compare our results with the other state-of-the-art FEM-based image registration methods. We study how the superimposed mesh structure would influence the Finite Element Method (FEM)-based image registration process.We propose a mesh generation algorithm based on how the mesh will influence the registration process, using the discrete Centroidal Voronoi Tessellation idea.We present a parallel algorithm to compute and update the mesh structure efficiently during image registration.

Surface Approximation via Asymptotic Optimal Geometric Partition

In this paper, we present a novel method on surface partition from the perspective of approximation theory. Different from previous shape proxies, the ellipsoidal variance proxy is proposed to penalize the partition results falling into disconnected parts. On its support, the Principle Component Analysis (PCA) based energy is developed for asymptotic cluster aspect ratio and size control. We provide the theoretical explanation on how the minimization of the PCA-based energy leads to the optimal asymptotic behavior for approximation. Moreover, we show the partitions on densely sampled triangular meshes converge to the theoretic expectations. To evaluate the effectiveness of surface approximation, polygonal/triangular surface remeshing results are generated. The experimental results demonstrate the high approximation quality of our method.

Robust 3D face modeling and reconstruction from frontal and side images

Robust and effective capture and reconstruction of 3D face models directly by smartphone users enables many applications. This paper presents a novel 3D face modeling and reconstruction solution that robustly and accurately acquire 3D face models from a couple of images captured by a single smartphone camera. Two selfie photos of a subject taken from the front and side are first used to guide our Non-Negative Matrix Factorization (NMF) induced part-based face model to iteratively reconstruct an initial 3D face of the subject. Then, an iterative detail updating method is applied to the initial generated 3D face to reconstruct facial details through optimizing lighting parameters and local depths. Our iterative 3D face reconstruction method permits fully automatic registration of a part-based face representation to the acquired face data and the detailed 2D/3D features to build a high-quality 3D face model. The NMF part-based face representation learned from a 3D face database facilitates effective global and adaptive local detail data fitting alternatively. Our system is flexible and it allows users to conduct the capture in any uncontrolled environment. We demonstrate the capability of our method by allowing users to capture and reconstruct their 3D faces by themselves. A deformable part-based 3D face representation for robust data fitting.An improved morphable model for 3D face reconstruction under constraints.A novel 3D face acquisition solution using a single smartphone camera.

Sliver-suppressing tetrahedral mesh optimization with gradient-based shape matching energy

In this paper, a novel shape matching energy is proposed to suppress slivers for tetrahedral mesh generation. Given a volumetric domain with a user-specified template (regular) simplex, the tetrahedral meshing problem is transformed into a shape matching formulation with a gradient-based energy, i.e., the gradient of linear shape function. It effectively inhibits small heights and suppresses all the badly-shaped tetrahedrons in tetrahedral meshes. The proposed approach iteratively optimizes vertex positions and mesh connectivity, and makes the simplices in the computed mesh as close as possible to the template simplex. We compare our results qualitatively and quantitatively with the state-of-the-art algorithm in tetrahedral meshing on extensive models using the standard measurement criteria.

Spectral Animation Compression

This paper presents a spectral approach to compress dynamic animation consisting of a sequence of homeomorphic manifold meshes. Our new approach directly compresses the field of deformation gradient defined on the surface mesh, by decomposing it into rigid-body motion (rotation) and non-rigid-body deformation (stretching) through polar decomposition. It is known that the rotation group has the algebraic topology of 3D ring, which is different from other operations like stretching. Thus we compress these two groups separately, by using Manifold Harmonics Transform to drop out their high-frequency details. Our experimental result shows that the proposed method achieves a good balance between the reconstruction quality and the compression ratio. We compare our results quantitatively with other existing approaches on animation compression, using standard measurement criteria.