Neural implicit surface reconstruction of freehand 3D ultrasound volume with geometric constraints.

Three-dimensional (3D) freehand ultrasound (US) is a widely used imaging modality that allows non-invasive imaging of medical anatomy without radiation exposure. Surface reconstruction of US volume is vital to acquire the accurate anatomical structures needed for modeling, registration, and visualization. However, traditional methods cannot produce a high-quality surface due to image noise. Despite improvements in smoothness, continuity, and resolution from deep learning approaches, research on surface reconstruction in freehand 3D US is still limited. This study introduces FUNSR, a self-supervised neural implicit surface reconstruction method to learn signed distance functions (SDFs) from US volumes. In particular, FUNSR iteratively learns the SDFs by moving the 3D queries sampled around volumetric point clouds to approximate the surface, guided by two novel geometric constraints: sign consistency constraint and on-surface constraint with adversarial learning. Our approach has been thoroughly evaluated across four datasets to demonstrate its adaptability to various anatomical structures, including a hip phantom dataset, two vascular datasets and one publicly available prostate dataset. We also show that smooth and continuous representations greatly enhance the visual appearance of US data. Furthermore, we highlight the potential of our method to improve segmentation performance, and its robustness to noise distribution and motion perturbation.

A unified generation-registration framework for improved MR-based CT synthesis in proton therapy.

BACKGROUND: The use of magnetic resonance (MR) imaging for proton therapy treatment planning is gaining attention as a highly effective method for guidance. At the core of this approach is the generation of computed tomography (CT) images from MR scans. However, the critical issue in this process is accurately aligning the MR and CT images, a task that becomes particularly challenging in frequently moving body areas, such as the head-and-neck. Misalignments in these images can result in blurred synthetic CT (sCT) images, adversely affecting the precision and effectiveness of the treatment planning. PURPOSE: This study introduces a novel network that cohesively unifies image generation and registration processes to enhance the quality and anatomical fidelity of sCTs derived from better-aligned MR images. METHODS: The approach synergizes a generation network (G) with a deformable registration network (R), optimizing them jointly in MR-to-CT synthesis. This goal is achieved by alternately minimizing the discrepancies between the generated/registered CT images and their corresponding reference CT counterparts. The generation network employs a UNet architecture, while the registration network leverages an implicit neural representation (INR) of the displacement vector fields (DVFs). We validated this method on a dataset comprising 60 head-and-neck patients, reserving 12 cases for holdout testing. RESULTS: Compared to the baseline Pix2Pix method with MAE 124.95 ± $\pm$ 30.74 HU, the proposed technique demonstrated 80.98 ± $\pm$ 7.55 HU. The unified translation-registration network produced sharper and more anatomically congruent outputs, showing superior efficacy in converting MR images to sCTs. Additionally, from a dosimetric perspective, the plan recalculated on the resulting sCTs resulted in a remarkably reduced discrepancy to the reference proton plans. CONCLUSIONS: This study conclusively demonstrates that a holistic MR-based CT synthesis approach, integrating both image-to-image translation and deformable registration, significantly improves the precision and quality of sCT generation, particularly for the challenging body area with varied anatomic changes between corresponding MR and CT.

Structure-aware diffusion for low-dose CT imaging.

Reducing the radiation dose leads to the x-ray computed tomography (CT) images suffering from heavy noise and artifacts, which inevitably interferes with the subsequent clinic diagnostic and analysis. Leading works have explored diffusion models for low-dose CT imaging to avoid the structure degeneration and blurring effects of previous deep denoising models. However, most of them always begin their generative processes with Gaussian noise, which has little or no structure priors of the clean data distribution, thereby leading to long-time inference and unpleasant reconstruction quality. To alleviate these problems, this paper presents a Structure-Aware Diffusion model (SAD), an end-to-end self-guided learning framework for high-fidelity CT image reconstruction. First, SAD builds a nonlinear diffusion bridge between clean and degraded data distributions, which could directly learn the implicit physical degradation prior from observed measurements. Second, SAD integrates the prompt learning mechanism and implicit neural representation into the diffusion process, where rich and diverse structure representations extracted by degraded inputs are exploited as prompts, which provides global and local structure priors, to guide CT image reconstruction. Finally, we devise an efficient self-guided diffusion architecture using an iterative updated strategy, which further refines structural prompts during each generative step to drive finer image reconstruction. Extensive experiments on AAPM-Mayo and LoDoPaB-CT datasets demonstrate that our SAD could achieve superior performance in terms of noise removal, structure preservation, and blind-dose generalization, with few generative steps, even one step only.

Dynamic CBCT imaging using prior model-free spatiotemporal implicit neural representation (PMF-STINR).

Objective. Dynamic cone-beam computed tomography (CBCT) can capture high-spatial-resolution, time-varying images for motion monitoring, patient setup, and adaptive planning of radiotherapy. However, dynamic CBCT reconstruction is an extremely ill-posed spatiotemporal inverse problem, as each CBCT volume in the dynamic sequence is only captured by one or a few x-ray projections, due to the slow gantry rotation speed and the fast anatomical motion (e.g. breathing).Approach. We developed a machine learning-based technique, prior-model-free spatiotemporal implicit neural representation (PMF-STINR), to reconstruct dynamic CBCTs from sequentially acquired x-ray projections. PMF-STINR employs a joint image reconstruction and registration approach to address the under-sampling challenge, enabling dynamic CBCT reconstruction from singular x-ray projections. Specifically, PMF-STINR uses spatial implicit neural representations to reconstruct a reference CBCT volume, and it applies temporal INR to represent the intra-scan dynamic motion of the reference CBCT to yield dynamic CBCTs. PMF-STINR couples the temporal INR with a learning-based B-spline motion model to capture time-varying deformable motion during the reconstruction. Compared with the previous methods, the spatial INR, the temporal INR, and the B-spline model of PMF-STINR are all learned on the fly during reconstruction in a one-shot fashion, without using any patient-specific prior knowledge or motion sorting/binning.Main results. PMF-STINR was evaluated via digital phantom simulations, physical phantom measurements, and a multi-institutional patient dataset featuring various imaging protocols (half-fan/full-fan, full sampling/sparse sampling, different energy and mAs settings, etc). The results showed that the one-shot learning-based PMF-STINR can accurately and robustly reconstruct dynamic CBCTs and capture highly irregular motion with high temporal (∼ 0.1 s) resolution and sub-millimeter accuracy.Significance. PMF-STINR can reconstruct dynamic CBCTs and solve the intra-scan motion from conventional 3D CBCT scans without using any prior anatomical/motion model or motion sorting/binning. It can be a promising tool for motion management by offering richer motion information than traditional 4D-CBCTs.

Iterative reconstruction for limited-angle CT using implicit neural representation.

Objective.Limited-angle computed tomography (CT) presents a challenge due to its ill-posed nature. In such scenarios, analytical reconstruction methods often exhibit severe artifacts. To tackle this inverse problem, several supervised deep learning-based approaches have been proposed. However, they are constrained by limitations such as generalization issue and the difficulty of acquiring a large amount of paired CT images.Approach.In this work, we propose an iterative neural reconstruction framework designed for limited-angle CT. By leveraging a coordinate-based neural representation, we formulate tomographic reconstruction as a convex optimization problem involving a deep neural network. We then employ differentiable projection layer to optimize this network by minimizing the discrepancy between the predicted and measured projection data. In addition, we introduce a prior-based weight initialization method to ensure the network starts optimization with an informed initial guess. This strategic initialization significantly improves the quality of iterative reconstruction by stabilizing the divergent behavior in ill-posed neural fields. Our method operates in a self-supervised manner, thereby eliminating the need for extensive data.Main results.The proposed method outperforms other iterative and learning-based methods. Experimental results on XCAT and Mayo Clinic datasets demonstrate the effectiveness of our approach in restoring anatomical features as well as structures. This finding was substantiated by visual inspections and quantitative evaluations using NRMSE, PSNR, and SSIM. Moreover, we conduct a comprehensive investigation into the divergent behavior of iterative neural reconstruction, thus revealing its suboptimal convergence when starting from scratch. In contrast, our method consistently produced accurate images by incorporating an initial estimate as informed initialization.Significance.This work showcases the feasibility to reconstruct high-fidelity CT images from limited-angle x-ray projections. The proposed methodology introduces a novel data-free approach to enhance medical imaging, holding promise across various clinical applications.

A subject-specific unsupervised deep learning method for quantitative susceptibility mapping using implicit neural representation.

Quantitative susceptibility mapping (QSM) is an MRI-based technique that estimates the underlying tissue magnetic susceptibility based on phase signal. Deep learning (DL)-based methods have shown promise in handling the challenging ill-posed inverse problem for QSM reconstruction. However, they require extensive paired training data that are typically unavailable and suffer from generalization problems. Recent model-incorporated DL approaches also overlook the non-local effect of the tissue phase in applying the source-to-field forward model due to patch-based training constraint, resulting in a discrepancy between the prediction and measurement and subsequently suboptimal QSM reconstruction. This study proposes an unsupervised and subject-specific DL method for QSM reconstruction based on implicit neural representation (INR), referred to as INR-QSM. INR has emerged as a powerful framework for learning a high-quality continuous representation of the signal (image) by exploiting its internal information without training labels. In INR-QSM, the desired susceptibility map is represented as a continuous function of the spatial coordinates, parameterized by a fully-connected neural network. The weights are learned by minimizing a loss function that includes a data fidelity term incorporated by the physical model and regularization terms. Additionally, a novel phase compensation strategy is proposed for the first time to account for the non-local effect of tissue phase in data consistency calculation to make the physical model more accurate. Our experiments show that INR-QSM outperforms traditional established QSM reconstruction methods and the compared unsupervised DL method both qualitatively and quantitatively, and is competitive against supervised DL methods under data perturbations.

Semi-supervised model based on implicit neural representation and mutual learning (SIMN) for multi-center nasopharyngeal carcinoma segmentation on MRI.

BACKGROUND: The issue of using deep learning to obtain accurate gross tumor volume (GTV) and metastatic lymph nodes (MLN) segmentation for nasopharyngeal carcinoma (NPC) on heterogeneous magnetic resonance imaging (MRI) images with limited labeling remains unsolved. METHOD: We collected 918 patients with MRI images from three hospitals to develop and validate models and proposed a semi-supervised framework for the fine delineation of multi-center NPC boundaries by integrating uncertainty-based implicit neural representations named SIMN. The framework utilizes the deep mutual learning approach with CNN and Transformer, incorporating dynamic thresholds. Additionally, domain adaptive algorithms are employed to enhance the performance. RESULTS: SIMN predictions have a high overlap ratio with the ground truth. Under the 20 % labeled cases, for the internal test cohorts, the average DSC in GTV and MLN are 0.7981 and 0.7804, respectively; for external test cohort Wu Zhou Red Cross Hospital, the average DSC in GTV and MLN are 0.7217 and 0.7581, respectively; for external test cohorts First People Hospital of Foshan, the average DSC in GTV and MLN are 0.7004 and 0.7692, respectively. No significant differences are found in DSC, HD95, ASD, and Recall for patients with different clinical categories. Moreover, SIMN outperformed existing classical semi-supervised methods. CONCLUSIONS: SIMN showed a highly accurate GTV and MLN segmentation for NPC on multi-center MRI images under Semi-Supervised Learning (SSL), which can easily transfer to other centers without fine-tuning. It suggests that it has the potential to act as a generalized delineation solution for heterogeneous MRI images with limited labels in clinical deployment.

3D cine-magnetic resonance imaging using spatial and temporal implicit neural representation learning (STINR-MR).

Objective. 3D cine-magnetic resonance imaging (cine-MRI) can capture images of the human body volume with high spatial and temporal resolutions to study anatomical dynamics. However, the reconstruction of 3D cine-MRI is challenged by highly under-sampled k-space data in each dynamic (cine) frame, due to the slow speed of MR signal acquisition. We proposed a machine learning-based framework, spatial and temporal implicit neural representation learning (STINR-MR), for accurate 3D cine-MRI reconstruction from highly under-sampled data.Approach. STINR-MR used a joint reconstruction and deformable registration approach to achieve a high acceleration factor for cine volumetric imaging. It addressed the ill-posed spatiotemporal reconstruction problem by solving a reference-frame 3D MR image and a corresponding motion model that deforms the reference frame to each cine frame. The reference-frame 3D MR image was reconstructed as a spatial implicit neural representation (INR) network, which learns the mapping from input 3D spatial coordinates to corresponding MR values. The dynamic motion model was constructed via a temporal INR, as well as basis deformation vector fields (DVFs) extracted from prior/onboard 4D-MRIs using principal component analysis. The learned temporal INR encodes input time points and outputs corresponding weighting factors to combine the basis DVFs into time-resolved motion fields that represent cine-frame-specific dynamics. STINR-MR was evaluated using MR data simulated from the 4D extended cardiac-torso (XCAT) digital phantom, as well as two MR datasets acquired clinically from human subjects. Its reconstruction accuracy was also compared with that of the model-based non-rigid motion estimation method (MR-MOTUS) and a deep learning-based method (TEMPEST).Main results. STINR-MR can reconstruct 3D cine-MR images with high temporal (<100 ms) and spatial (3 mm) resolutions. Compared with MR-MOTUS and TEMPEST, STINR-MR consistently reconstructed images with better image quality and fewer artifacts and achieved superior tumor localization accuracy via the solved dynamic DVFs. For the XCAT study, STINR reconstructed the tumors to a mean ± SD center-of-mass error of 0.9 ± 0.4 mm, compared to 3.4 ± 1.0 mm of the MR-MOTUS method. The high-frame-rate reconstruction capability of STINR-MR allows different irregular motion patterns to be accurately captured.Significance. STINR-MR provides a lightweight and efficient framework for accurate 3D cine-MRI reconstruction. It is a 'one-shot' method that does not require external data for pre-training, allowing it to avoid generalizability issues typically encountered in deep learning-based methods.

Self-supervised multicontrast super-resolution for diffusion-weighted prostate MRI.

PURPOSE: This study addresses the challenge of low resolution and signal-to-noise ratio (SNR) in diffusion-weighted images (DWI), which are pivotal for cancer detection. Traditional methods increase SNR at high b-values through multiple acquisitions, but this results in diminished image resolution due to motion-induced variations. Our research aims to enhance spatial resolution by exploiting the global structure within multicontrast DWI scans and millimetric motion between acquisitions. METHODS: We introduce a novel approach employing a "Perturbation Network" to learn subvoxel-size motions between scans, trained jointly with an implicit neural representation (INR) network. INR encodes the DWI as a continuous volumetric function, treating voxel intensities of low-resolution acquisitions as discrete samples. By evaluating this function with a finer grid, our model predicts higher-resolution signal intensities for intermediate voxel locations. The Perturbation Network's motion-correction efficacy was validated through experiments on biological phantoms and in vivo prostate scans. RESULTS: Quantitative analyses revealed significantly higher structural similarity measures of super-resolution images to ground truth high-resolution images compared to high-order interpolation (p < $$ < $$ 0.005). In blind qualitative experiments, 96 . 1 % $$ 96.1\% $$ of super-resolution images were assessed to have superior diagnostic quality compared to interpolated images. CONCLUSION: High-resolution details in DWI can be obtained without the need for high-resolution training data. One notable advantage of the proposed method is that it does not require a super-resolution training set. This is important in clinical practice because the proposed method can easily be adapted to images with different scanner settings or body parts, whereas the supervised methods do not offer such an option.

Arbitrary scale super-resolution diffusion model for brain MRI images.

Given the constraints posed by hardware capacity, scan duration, and patient cooperation, the reconstruction of magnetic resonance imaging (MRI) images emerges as a pivotal aspect of medical imaging research. Currently, deep learning-based super-resolution (SR) methods have been widely discussed in medical image processing due to their ability to reconstruct high-quality, high resolution (HR) images from low resolution (LR) inputs. However, most existing MRI SR methods are designed for specific magnifications and cannot generate MRI images at arbitrary scales, which hinders the radiologists from fully visualizing the lesions. Moreover, current arbitrary scale SR methods often suffer from issues like excessive smoothing and artifacts. In this paper, we propose an Arbitrary Scale Super-Resolution Diffusion Model (ASSRDM), which combines implicit neural representation with the denoising diffusion probabilistic model to achieve arbitrary-scale, high-fidelity medical images SR. Moreover, we formulate a continuous resolution regulation mechanism, comprising a multi-scale LR guidance network and a scaling factor. The scaling factor finely adjusts the resolution and dynamically influences the weighting of LR details and synthesized features in the final output. This capability allows the model to seamlessly adapt to the requirements of continuous resolution adjustments. Additionally, the multi-scale LR guidance network provides the denoising block with multi-resolution LR features to enrich texture information and restore high-frequency details. Extensive experiments conducted on the IXI and fastMRI datasets demonstrate that our ASSRDM exhibits superior performance compared to existing techniques and has tremendous potential in clinical practice.

Generative modeling of living cells with SO(3)-equivariant implicit neural representations.

Data-driven cell tracking and segmentation methods in biomedical imaging require diverse and information-rich training data. In cases where the number of training samples is limited, synthetic computer-generated data sets can be used to improve these methods. This requires the synthesis of cell shapes as well as corresponding microscopy images using generative models. To synthesize realistic living cell shapes, the shape representation used by the generative model should be able to accurately represent fine details and changes in topology, which are common in cells. These requirements are not met by 3D voxel masks, which are restricted in resolution, and polygon meshes, which do not easily model processes like cell growth and mitosis. In this work, we propose to represent living cell shapes as level sets of signed distance functions (SDFs) which are estimated by neural networks. We optimize a fully-connected neural network to provide an implicit representation of the SDF value at any point in a 3D+time domain, conditioned on a learned latent code that is disentangled from the rotation of the cell shape. We demonstrate the effectiveness of this approach on cells that exhibit rapid deformations (Platynereis dumerilii), cells that grow and divide (C. elegans), and cells that have growing and branching filopodial protrusions (A549 human lung carcinoma cells). A quantitative evaluation using shape features and Dice similarity coefficients of real and synthetic cell shapes shows that our model can generate topologically plausible complex cell shapes in 3D+time with high similarity to real living cell shapes. Finally, we show how microscopy images of living cells that correspond to our generated cell shapes can be synthesized using an image-to-image model.

OW-SLR: Overlapping Windows on Semi-Local Region for Image Super-Resolution.

There has been considerable progress in implicit neural representation to upscale an image to any arbitrary resolution. However, existing methods are based on defining a function to predict the Red, Green and Blue (RGB) value from just four specific loci. Relying on just four loci is insufficient as it leads to losing fine details from the neighboring region(s). We show that by taking into account the semi-local region leads to an improvement in performance. In this paper, we propose applying a new technique called Overlapping Windows on Semi-Local Region (OW-SLR) to an image to obtain any arbitrary resolution by taking the coordinates of the semi-local region around a point in the latent space. This extracted detail is used to predict the RGB value of a point. We illustrate the technique by applying the algorithm to the Optical Coherence Tomography-Angiography (OCT-A) images and show that it can upscale them to random resolution. This technique outperforms the existing state-of-the-art methods when applied to the OCT500 dataset. OW-SLR provides better results for classifying healthy and diseased retinal images such as diabetic retinopathy and normals from the given set of OCT-A images.

Dynamic CBCT Imaging using Prior Model-Free Spatiotemporal Implicit Neural Representation (PMF-STINR).

Objective: Dynamic cone-beam computed tomography (CBCT) can capture high-spatial-resolution, time-varying images for motion monitoring, patient setup, and adaptive planning of radiotherapy. However, dynamic CBCT reconstruction is an extremely ill-posed spatiotemporal inverse problem, as each CBCT volume in the dynamic sequence is only captured by one or a few X-ray projections, due to the slow gantry rotation speed and the fast anatomical motion (e.g., breathing). Approach: We developed a machine learning-based technique, prior-model-free spatiotemporal implicit neural representation (PMF-STINR), to reconstruct dynamic CBCTs from sequentially acquired X-ray projections. PMF-STINR employs a joint image reconstruction and registration approach to address the under-sampling challenge, enabling dynamic CBCT reconstruction from singular X-ray projections. Specifically, PMF-STINR uses spatial implicit neural representation to reconstruct a reference CBCT volume, and it applies temporal INR to represent the intra-scan dynamic motion with respect to the reference CBCT to yield dynamic CBCTs. PMF-STINR couples the temporal INR with a learning-based B-spline motion model to capture time-varying deformable motion during the reconstruction. Compared with the previous methods, the spatial INR, the temporal INR, and the B-spline model of PMF-STINR are all learned on the fly during reconstruction in a one-shot fashion, without using any patient-specific prior knowledge or motion sorting/binning. Main results: PMF-STINR was evaluated via digital phantom simulations, physical phantom measurements, and a multi-institutional patient dataset featuring various imaging protocols (half-fan/full-fan, full sampling/sparse sampling, different energy and mAs settings, etc.). The results showed that the one-shot learning-based PMF-STINR can accurately and robustly reconstruct dynamic CBCTs and capture highly irregular motion with high temporal (~0.1s) resolution and sub-millimeter accuracy. Significance: PMF-STINR can reconstruct dynamic CBCTs and solve the intra-scan motion from conventional 3D CBCT scans without using any prior anatomical/motion model or motion sorting/binning. It can be a promising tool for motion management by offering richer motion information than traditional 4D-CBCTs.

Super-resolution biomedical imaging via reference-free statistical implicit neural representation.

Objective.Supervised deep learning for image super-resolution (SR) has limitations in biomedical imaging due to the lack of large amounts of low- and high-resolution image pairs for model training. In this work, we propose a reference-free statistical implicit neural representation (INR) framework, which needs only a single or a few observed low-resolution (LR) image(s), to generate high-quality SR images.Approach.The framework models the statistics of the observed LR images via maximum likelihood estimation and trains the INR network to represent the latent high-resolution (HR) image as a continuous function in the spatial domain. The INR network is constructed as a coordinate-based multi-layer perceptron, whose inputs are image spatial coordinates and outputs are corresponding pixel intensities. The trained INR not only constrains functional smoothness but also allows an arbitrary scale in SR imaging.Main results.We demonstrate the efficacy of the proposed framework on various biomedical images, including computed tomography (CT), magnetic resonance imaging (MRI), fluorescence microscopy, and ultrasound images, across different SR magnification scales of 2×, 4×, and 8×. A limited number of LR images were used for each of the SR imaging tasks to show the potential of the proposed statistical INR framework.Significance.The proposed method provides an urgently needed unsupervised deep learning framework for numerous biomedical SR applications that lack HR reference images.

NRVC: Neural Representation for Video Compression with Implicit Multiscale Fusion Network.

Recently, end-to-end deep models for video compression have made steady advancements. However, this resulted in a lengthy and complex pipeline containing numerous redundant parameters. The video compression approaches based on implicit neural representation (INR) allow videos to be directly represented as a function approximated by a neural network, resulting in a more lightweight model, whereas the singularity of the feature extraction pipeline limits the network's ability to fit the mapping function for video frames. Hence, we propose a neural representation approach for video compression with an implicit multiscale fusion network (NRVC), utilizing normalized residual networks to improve the effectiveness of INR in fitting the target function. We propose the multiscale representations for video compression (MSRVC) network, which effectively extracts features from the input video sequence to enhance the degree of overfitting in the mapping function. Additionally, we propose the feature extraction channel attention (FECA) block to capture interaction information between different feature extraction channels, further improving the effectiveness of feature extraction. The results show that compared to the NeRV method with similar bits per pixel (BPP), NRVC has a 2.16% increase in the decoded peak signal-to-noise ratio (PSNR). Moreover, NRVC outperforms the conventional HEVC in terms of PSNR.

3D cine-magnetic resonance imaging using spatial and temporal implicit neural representation learning (STINR-MR).

Objective: 3D cine-magnetic resonance imaging (cine-MRI) can capture images of the human body volume with high spatial and temporal resolutions to study the anatomical dynamics. However, the reconstruction of 3D cine-MRI is challenged by highly undersampled k-space data in each dynamic (cine) frame, due to the slow speed of MR signal acquisition. We proposed a machine learning-based framework, spatial and temporal implicit neural representation learning (STINR-MR), for accurate 3D cine-MRI reconstruction from highly undersampled data. Approach: STINR-MR used a joint reconstruction and deformable registration approach to achieve a high acceleration factor for cine volumetric imaging. It addressed the ill-posed spatiotemporal reconstruction problem by solving a reference-frame 3D MR image and a corresponding motion model which deforms the reference frame to each cine frame. The reference-frame 3D MR image was reconstructed as a spatial implicit neural representation (INR) network, which learns the mapping from input 3D spatial coordinates to corresponding MR values. The dynamic motion model was constructed via a temporal INR, as well as basis deformation vector fields (DVFs) extracted from prior/onboard 4D-MRIs using principal component analysis (PCA). The learned temporal INR encodes input time points and outputs corresponding weighting factors to combine the basis DVFs into time-resolved motion fields that represent cine-frame-specific dynamics. STINR-MR was evaluated using MR data simulated from the 4D extended cardiac-torso (XCAT) digital phantom, as well as MR data acquired clinically from a healthy human subject. Its reconstruction accuracy was also compared with that of the model-based non-rigid motion estimation method (MR-MOTUS). Main results: STINR-MR can reconstruct 3D cine-MR images with high temporal (<100 ms) and spatial (3 mm) resolutions. Compared with MR-MOTUS, STINR-MR consistently reconstructed images with better image quality and fewer artifacts and achieved superior tumor localization accuracy via the solved dynamic DVFs. For the XCAT study, STINR reconstructed the tumors to a mean±S.D. center-of-mass error of 1.0±0.4 mm, compared to 3.4±1.0 mm of the MR-MOTUS method. The high-frame-rate reconstruction capability of STINR-MR allows different irregular motion patterns to be accurately captured. Significance: STINR-MR provides a lightweight and efficient framework for accurate 3D cine-MRI reconstruction. It is a 'one-shot' method that does not require external data for pre-training, allowing it to avoid generalizability issues typically encountered in deep learning-based methods.

SR-FEINR: Continuous Remote Sensing Image Super-Resolution Using Feature-Enhanced Implicit Neural Representation.

Remote sensing images often have limited resolution, which can hinder their effectiveness in various applications. Super-resolution techniques can enhance the resolution of remote sensing images, and arbitrary resolution super-resolution techniques provide additional flexibility in choosing appropriate image resolutions for different tasks. However, for subsequent processing, such as detection and classification, the resolution of the input image may vary greatly for different methods. In this paper, we propose a method for continuous remote sensing image super-resolution using feature-enhanced implicit neural representation (SR-FEINR). Continuous remote sensing image super-resolution means users can scale a low-resolution image into an image with arbitrary resolution. Our algorithm is composed of three main components: a low-resolution image feature extraction module, a positional encoding module, and a feature-enhanced multi-layer perceptron module. We are the first to apply implicit neural representation in a continuous remote sensing image super-resolution task. Through extensive experiments on two popular remote sensing image datasets, we have shown that our SR-FEINR outperforms the state-of-the-art algorithms in terms of accuracy. Our algorithm showed an average improvement of 0.05 dB over the existing method on ×30 across three datasets.

Dynamic cone-beam CT reconstruction using spatial and temporal implicit neural representation learning (STINR).

Objective. Dynamic cone-beam CT (CBCT) imaging is highly desired in image-guided radiation therapy to provide volumetric images with high spatial and temporal resolutions to enable applications including tumor motion tracking/prediction and intra-delivery dose calculation/accumulation. However, dynamic CBCT reconstruction is a substantially challenging spatiotemporal inverse problem, due to the extremely limited projection sample available for each CBCT reconstruction (one projection for one CBCT volume).Approach. We developed a simultaneous spatial and temporal implicit neural representation (STINR) method for dynamic CBCT reconstruction. STINR mapped the unknown image and the evolution of its motion into spatial and temporal multi-layer perceptrons (MLPs), and iteratively optimized the neuron weightings of the MLPs via acquired projections to represent the dynamic CBCT series. In addition to the MLPs, we also introduced prior knowledge, in the form of principal component analysis (PCA)-based patient-specific motion models, to reduce the complexity of the temporal mapping to address the ill-conditioned dynamic CBCT reconstruction problem. We used the extended-cardiac-torso (XCAT) phantom and a patient 4D-CBCT dataset to simulate different lung motion scenarios to evaluate STINR. The scenarios contain motion variations including motion baseline shifts, motion amplitude/frequency variations, and motion non-periodicity. The XCAT scenarios also contain inter-scan anatomical variations including tumor shrinkage and tumor position change.Main results. STINR shows consistently higher image reconstruction and motion tracking accuracy than a traditional PCA-based method and a polynomial-fitting-based neural representation method. STINR tracks the lung target to an average center-of-mass error of 1-2 mm, with corresponding relative errors of reconstructed dynamic CBCTs around 10%.Significance. STINR offers a general framework allowing accurate dynamic CBCT reconstruction for image-guided radiotherapy. It is a one-shot learning method that does not rely on pre-training and is not susceptible to generalizability issues. It also allows natural super-resolution. It can be readily applied to other imaging modalities as well.

Implicit Anatomical Rendering for Medical Image Segmentation with Stochastic Experts.

Integrating high-level semantically correlated contents and low-level anatomical features is of central importance in medical image segmentation. Towards this end, recent deep learning-based medical segmentation methods have shown great promise in better modeling such information. However, convolution operators for medical segmentation typically operate on regular grids, which inherently blur the high-frequency regions, i.e., boundary regions. In this work, we propose MORSE, a generic implicit neural rendering framework designed at an anatomical level to assist learning in medical image segmentation. Our method is motivated by the fact that implicit neural representation has been shown to be more effective in fitting complex signals and solving computer graphics problems than discrete grid-based representation. The core of our approach is to formulate medical image segmentation as a rendering problem in an end-to-end manner. Specifically, we continuously align the coarse segmentation prediction with the ambiguous coordinate-based point representations and aggregate these features to adaptively refine the boundary region. To parallelly optimize multi-scale pixel-level features, we leverage the idea from Mixture-of-Expert (MoE) to design and train our MORSE with a stochastic gating mechanism. Our experiments demonstrate that MORSE can work well with different medical segmentation backbones, consistently achieving competitive performance improvements in both 2D and 3D supervised medical segmentation methods. We also theoretically analyze the superiority of MORSE.

Implicit neural representation for radiation therapy dose distribution.

Objective. Dose distribution data plays a pivotal role in radiotherapy treatment planning. The data is typically represented using voxel grids, and its size ranges from 106to 108. A concise representation of the treatment plan is of great value in facilitating treatment planning and downstream applications. This work aims to develop an implicit neural representation of 3D dose distribution data.Approach. Instead of storing the dose values at each voxel, in the proposed approach, the weights of a multilayer perceptron (MLP) are employed to characterize the dosimetric data for plan representation and subsequent applications. We train a coordinate-based MLP with sinusoidal activations to map the voxel spatial coordinates to the corresponding dose values. We identify the best architecture for a given parameter budget and use that to train a model for each patient. The trained MLP is evaluated at each voxel location to reconstruct the dose distribution. We perform extensive experiments on dose distributions of prostate, spine, and head and neck tumor cases to evaluate the quality of the proposed representation. We also study the change in representation quality by varying model size and activation function.Main results. Using coordinate-based MLPs with sinusoidal activations, we can learn implicit representations that achieve a mean-squared error of 10-6and peak signal-to-noise ratio greater than 50 dB at a target bitrate of ∼1 across all the datasets, with a compression ratio of ∼32. Our results also show that model sizes with a bitrate of 1-2 achieve optimal accuracy. For smaller bitrates, performance starts to drop significantly.Significance. The proposed model provides a low-dimensional, implicit, and continuous representation of 3D dose data. In summary, given a dose distribution, we systematically show how to find a compact model to fit the data accurately. This study lays the groundwork for future applications of neural representations of dose data in radiation oncology.