High Frequency Sampling of TTL Pulses on a Raspberry Pi for Diffuse Correlation Spectroscopy Applications.

Diffuse Correlation Spectroscopy (DCS) is a well-established optical technique that has been used for non-invasive measurement of blood flow in tissues. Instrumentation for DCS includes a correlation device that computes the temporal intensity autocorrelation of a coherent laser source after it has undergone diffuse scattering through a turbid medium. Typically, the signal acquisition and its autocorrelation are performed by a correlation board. These boards have dedicated hardware to acquire and compute intensity autocorrelations of rapidly varying input signal and usually are quite expensive. Here we show that a Raspberry Pi minicomputer can acquire and store a rapidly varying time-signal with high fidelity. We show that this signal collected by a Raspberry Pi device can be processed numerically to yield intensity autocorrelations well suited for DCS applications. DCS measurements made using the Raspberry Pi device were compared to those acquired using a commercial hardware autocorrelation board to investigate the stability, performance, and accuracy of the data acquired in controlled experiments. This paper represents a first step toward lowering the instrumentation cost of a DCS system and may offer the potential to make DCS become more widely used in biomedical applications.

Diffusion Posterior Sampling for Nonlinear CT Reconstruction.

Diffusion models have been demonstrated as powerful deep learning tools for image generation in CT reconstruction and restoration. Recently, diffusion posterior sampling, where a score-based diffusion prior is combined with a likelihood model, has been used to produce high quality CT images given low-quality measurements. This technique is attractive since it permits a one-time, unsupervised training of a CT prior; which can then be incorporated with an arbitrary data model. However, current methods only rely on a linear model of x-ray CT physics to reconstruct or restore images. While it is common to linearize the transmission tomography reconstruction problem, this is an approximation to the true and inherently nonlinear forward model. We propose a new method that solves the inverse problem of nonlinear CT image reconstruction via diffusion posterior sampling. We implement a traditional unconditional diffusion model by training a prior score function estimator, and apply Bayes rule to combine this prior with a measurement likelihood score function derived from the nonlinear physical model to arrive at a posterior score function that can be used to sample the reverse-time diffusion process. This plug-and-play method allows incorporation of a diffusion-based prior with generalized nonlinear CT image reconstruction into multiple CT system designs with different forward models, without the need for any additional training. We demonstrate the technique in both fully sampled low dose data and sparse-view geometries using a single unsupervised training of the prior.

Design Optimization of A Triple-Layer Flat-Panel Detector for Three-Material Decomposition.

Spectral radiography and fluoroscopy with multi-layer flat-panel detectors (FPD) is being actively investigated in a range of clinical applications. For applications involving contrast administration, maximal contrast resolution is achieved when overlaying background anatomy is completely removed. This calls for three-material decomposition (soft tissue, bone, and contrast) enabled by measurements in three energy channels. We have previously demonstrated the feasibility of such decomposition using a triple-layer detector. While algorithmic solutions can be adopted to mitigate noise in the material basis images, in this work, we seek to fundamentally improve the conditioning of the problem through optimized system design. Design parameters include source voltage, the thickness of the top two CsI scintillators, and the thickness of two copper interstitial filters. The design objective is to minimize noise in the basis image containing contrast, chosen as gadolinium in this work to improve separation from bone. The optimized design was compared with other designs with unoptimized scintillator thickness and/or without interstitial filtration. Results show that CsI thickness optimization and interstitial filtration can significantly reduce noise in the gadolinium image by 35.7% and 42.7% respectively within a lung ROI, which in turn boosts detectability of small vessels. Gadolinium and bone signals are separated in all cases. Visualization of coronary vessels is enabled by the combining optimized system design and regularization. Results from this work demonstrate that three-material decomposition can be significantly improved with system design optimization. Optimized designs obtained from this work can inform imaging techniques selection and triple-layer detector fabrication for spectral radiography.

CT reconstruction using diffusion posterior sampling conditioned on a nonlinear measurement model.

Purpose: Recently, diffusion posterior sampling (DPS), where score-based diffusion priors are combined with likelihood models, has been used to produce high-quality computed tomography (CT) images given low-quality measurements. This technique permits one-time, unsupervised training of a CT prior, which can then be incorporated with an arbitrary data model. However, current methods rely on a linear model of X-ray CT physics to reconstruct. Although it is common to linearize the transmission tomography reconstruction problem, this is an approximation to the true and inherently nonlinear forward model. We propose a DPS method that integrates a general nonlinear measurement model. Approach: We implement a traditional unconditional diffusion model by training a prior score function estimator and apply Bayes' rule to combine this prior with a measurement likelihood score function derived from the nonlinear physical model to arrive at a posterior score function that can be used to sample the reverse-time diffusion process. We develop computational enhancements for the approach and evaluate the reconstruction approach in several simulation studies. Results: The proposed nonlinear DPS provides improved performance over traditional reconstruction methods and DPS with a linear model. Moreover, as compared with a conditionally trained deep learning approach, the nonlinear DPS approach shows a better ability to provide high-quality images for different acquisition protocols. Conclusion: This plug-and-play method allows the incorporation of a diffusion-based prior with a general nonlinear CT measurement model. This permits the application of the approach to different systems, protocols, etc., without the need for any additional training.

CT Reconstruction using Diffusion Posterior Sampling conditioned on a Nonlinear Measurement Model.

Diffusion models have been demonstrated as powerful deep learning tools for image generation in CT reconstruction and restoration. Recently, diffusion posterior sampling, where a score-based diffusion prior is combined with a likelihood model, has been used to produce high quality CT images given low-quality measurements. This technique is attractive since it permits a one-time, unsupervised training of a CT prior; which can then be incorporated with an arbitrary data model. However, current methods rely on a linear model of x-ray CT physics to reconstruct or restore images. While it is common to linearize the transmission tomography reconstruction problem, this is an approximation to the true and inherently nonlinear forward model. We propose a new method that solves the inverse problem of nonlinear CT image reconstruction via diffusion posterior sampling. We implement a traditional unconditional diffusion model by training a prior score function estimator, and apply Bayes rule to combine this prior with a measurement likelihood score function derived from the nonlinear physical model to arrive at a posterior score function that can be used to sample the reverse-time diffusion process. This plug-and-play method allows incorporation of a diffusion-based prior with generalized nonlinear CT image reconstruction into multiple CT system designs with different forward models, without the need for any additional training. We develop the algorithm that performs this reconstruction, including an ordered-subsets variant for accelerated processing and demonstrate the technique in both fully sampled low dose data and sparse-view geometries using a single unsupervised training of the prior.

Deep Learning CT Image Restoration using System Blur Models.

Restoration of images contaminated by blur is an important processing tool across modalities including computed tomography where the blur induced by various system factors can be complex with dependencies on acquisition and reconstruction protocol, and even be patient-dependent. In many cases, such a blur can be modeled and predicted with high accuracy providing an important input to a classical deconvolution approach. While traditional deblurring methods tend to be highly noise magnifying, deep learning approaches have the potential to improve upon classic performance limits. However, most network architectures base their restoration on data inputs alone without knowledge of the system blur. In this work, we explore a deep learning approach that takes both image inputs as well as information that characterizes the system blur to combine modeling and deep learning approaches. We apply the approach to CT image restoration and compare with an image-only deep learning approach. We find that inclusion of the system blur model improves deblurring performance - suggesting the potential power of the combined modeling and deep learning technique.

Tunable neural networks for CT image formation.

Optimization of CT image quality typically involves balancing variance and bias. In traditional filtered back-projection, this trade-off is controlled by the filter cutoff frequency. In model-based iterative reconstruction, the regularization strength parameter often serves the same function. Deep neural networks (DNNs) typically do not provide this tunable control over output image properties. Models are often trained to minimize the expected mean squared error, which penalizes both variance and bias in image outputs but does not offer any control over the trade-off between the two. We propose a method for controlling the output image properties of neural networks with a new loss function called weighted covariance and bias (WCB). Our proposed method uses multiple noise realizations of the input images during training to allow for separate weighting matrices for the variance and bias penalty terms. Moreover, we show that tuning these weights enables targeted penalization of specific image features with spatial frequency domain penalties. To evaluate our method, we present a simulation study using digital anthropomorphic phantoms, physical simulation of CT measurements, and image formation with various algorithms. We show that the WCB loss function offers a greater degree of control over trade-offs between variance and bias, whereas mean-squared error provides only one specific image quality configuration. We also show that WCB can be used to control specific image properties including variance, bias, spatial resolution, and the noise correlation of neural network outputs. Finally, we present a method to optimize the proposed weights for a spiculated lung nodule shape discrimination task. Our results demonstrate this new image quality can control the image properties of DNN outputs and optimize image quality for task-specific performance.

Task-Driven CT Image Quality Optimization for Low-Contrast Lesion Detectability with Tunable Neural Networks.

Low-contrast lesions are difficult to detect in noisy low-dose CT images. Improving CT image quality for this detection task has the potential to improve diagnostic accuracy and patient outcomes. In this work, we use tunable neural networks for CT image restoration with a hyperparameter to control the variance/bias tradeoff. We use clinical images from a super-high-resolution normal-dose CT scan to synthesize low-contrast low-dose CT images for supervised training of deep learning CT reconstruction models. Those models are trained using with multiple noise realizations so that variance and bias can be penalized separately. We use a training loss function with one hyperparameter called the denoising level, which controls the variance/bias tradeoff. Finally, we evaluate the CT image quality to find the optimal denoising level for low-contrast lesion detectability. We evaluate performance using a shallow neural network model classification model to represent a suboptimal image observer. Our results indicate that the optimal networks for low-contrast lesion detectability are those that prioritize bias reduction rather than mean-squared error, which demonstrates the potential clinical benefit of our proposed tunable neural networks.

Notice of Removal: Fourier Diffusion Models: A Method to Control MTF and NPS in Score-Based Stochastic Image Generation.

Removed.

Design Optimization of Spatial-Spectral Filters for Cone-Beam CT Material Decomposition.

Spectral CT has shown promise for high-sensitivity quantitative imaging and material decomposition. This work presents a new device called a spatial-spectral filter (SSF) which consists of a tiled array of filter materials positioned near the x-ray source that is used to modulate the spectral shape of the x-ray beam. The filter is moved to obtain projection data that is sparse in each spectral channel. To process this sparse data, we employ a one-step direct model-based material decomposition (MBMD) to reconstruct basis material density images directly from the SSF CT data. To evaluate different possible SSF designs, we define a new Fisher-information-based predictive image quality metric called separability index which characterizes the ability of a spectral CT system to distinguish between the signals from two or more materials. This spectral CT performance metric can be used to optimize spectral CT system design. We conducted simulation-based design optimization study to find optimized combinations of filter materials, filter thicknesses, filter widths, and source settings. Finally, we present MBMD results using simulated SSF CT measurements from the optimized designs to demonstrate the ability to reconstruct basis material density images and to show the benefits of the optimized designs. Our results indicate that optimizing SSF CT for separability leads to high-performance at material discrimination tasks.

Control of Variance and Bias in CT Image Processing with Variational Training of Deep Neural Networks.

Optimization of CT image quality typically involves balancing noise and bias. In filtered back-projection, this trade-off is controlled by the particular filter and cutoff frequency. In penalized-likelihood iterative reconstruction, the penalty weight serves the same function. Deep neural networks typically do not provide this tuneable control over output image properties. Models are often trained to minimize mean squared error which penalizes both variance and bias in image outputs but does not offer any control over the trade-off between the two. In this work, we propose a method for controlling the output image properties of neural networks with a new loss function called weighted covariance and bias (WCB). Our proposed method includes separate weighting parameters to control the relative importance of noise or bias reduction. Moreover, we show that tuning these weights enables targeted penalization of specific image features (e.g. spatial frequencies). To evaluate our method, we present a simulation study using digital anthropormorphic phantoms, physical simulation of non-ideal CT data, and image formation with various algorithms. We show that WCB offers a greater degree of control over trade-offs between variance and bias whereas MSE has only one configuration. We also show that WCB can be used to control specific image properties including variance, bias, spatial resolution, and the noise correlation of neural network outputs. Finally, we present a method to optimize the proposed weights for stimulus detectability. Our results demonstrate the potential for this new capability to control the image properties of DNN outputs and optimize image quality for the task-specific applications.

Data-dependent Nonlinearity Analysis in CT Denoising CNNs.

Recent years have seen the increasing application of deep learning methods in medical imaging formation, processing, and analysis. These methods take advantage of the flexibility of nonlinear neural network models to encode information and features in ways that can outperform conventional approaches. However, because of the nonlinear nature of this processing, images formed by neural networks have properties that are highly data-dependent and difficult to analyze. In particular, the generalizability and robustness of these approaches can be difficult to ascertain. In this work, we analyze a class of neural networks that use only piece-wise linear activation functions. This class of networks can be represented by locally linear systems where the linear properties are highly data-dependent - allowing, for example, estimation of noise in image output via standard propagation methods. We propose a nonlinearity index metric that quantifies the fidelity of a local linear approximation of trained models based on specific input data. We applied this analysis to three example CT denoising CNNs to analytically predict the noise properties in the output images. We found that the proposed nonlinearity metric highly correlates with the accuracy of noise predictions. The analysis proposed in this work provides theoretical understanding of the nonlinear behavior of neural networks and enables performance prediction and quantitation under certain conditions.

Performance Assessment Framework for Neural Network Denoising.

The proliferation of deep learning image processing calls for a quantitative image quality assessment framework that is suitable for nonlinear, data-dependent algorithms. In this work, we propose a method to systematically evaluate the system and noise responses such that the nonlinear transfer properties can be mapped out. The method involves sampling of lesion perturbations as a function of size, contrast, as well as clinically relevant features such as shape and texture that may be important for diagnosis. We embed the perturbations in backgrounds of varying attenuation levels, noise magnitude and correlation that are associated with different patient anatomies and imaging protocols. The range of system and noise response are further used to evaluate performance for clinical tasks such as signal detection and classification. We performed the assessment for an example CNN-denoising algorithm for low does lung CT screening. The system response of the CNN-denoising algorithm exhibits highly nonlinear behavior where both contrast and higher order lesion features such as spiculated boundaries are not reliably represented for lesions perturbations with small size and low contrast. The noise properties are potentially highly nonstationary, and should be assumed to be the same between the signal-present and signal-absent images. Furthermore, we observer a high degree dependency of both system and noise response on the background attenuation levels. Inputs around zeros are effectively imposed a non-negativity constraint; transfer properties for higher background levels are highly variable. For a detection task, CNN-denoised images improved detectability index by 16-18% compared to low dose CT inputs. For classification task between spiculated and smooth lesions, CNN-denoised images result in a much larger improvement up to 50%. The performance assessment framework propose in this work can systematically map out the nonlinear transfer functions for deep learning algorithms and can potentially enable robust deployment of such algorithms in medical imaging applications.

Model-based three-material decomposition in dual-energy CT using the volume conservation constraint.

Objective. We develop a model-based optimization algorithm for 'one-step' dual-energy (DE) CT decomposition of three materials directly from projection measurements.Approach.Since the three-material problem is inherently undetermined, we incorporate the volume conservation principle (VCP) as a pair of equality and nonnegativity constraints into the objective function of the recently reported model-based material decomposition (MBMD). An optimization algorithm (constrained MBMD, CMBMD) is derived that utilizes voxel-wise separability to partition the volume into a VCP-constrained region solved using interior-point iterations, and an unconstrained region (air surrounding the object, where VCP is violated) solved with conventional two-material MBMD. Constrained MBMD (CMBMD) is validated in simulations and experiments in application to bone composition measurements in the presence of metal hardware using DE cone-beam CT (CBCT). A kV-switching protocol with non-coinciding low- and high-energy (LE and HE) projections was assumed. CMBMD with decomposed base materials of cortical bone, fat, and metal (titanium, Ti) is compared to MBMD with (i) fat-bone and (ii) fat-Ti bases.Main results.Three-material CMBMD exhibits a substantial reduction in metal artifacts relative to the two-material MBMD implementations. The accuracies of cortical bone volume fraction estimates are markedly improved using CMBMD, with ∼5-10× lower normalized root mean squared error in simulations with anthropomorphic knee phantoms (depending on the complexity of the metal component) and ∼2-2.5× lower in an experimental test-bench study.Significance.In conclusion, we demonstrated one-step three-material decomposition of DE CT using volume conservation as an optimization constraint. The proposed method might be applicable to DE applications such as bone marrow edema imaging (fat-bone-water decomposition) or multi-contrast imaging, especially on CT/CBCT systems that do not provide coinciding LE and HE ray paths required for conventional projection-domain DE decomposition.

Tunable Neural Networks for Multi-Material Image Formation from Spectral CT Measurements.

Quantitative estimation of multi-material density images is an important goal for Spectral CT imaging. However, material decomposition is a poorly-conditioned nonlinear inverse problem. Maximum-likelihood model-based material decomposition results in very noisy material density image estimates. One increasingly popular strategy for noise reduction is to apply deep neural networks for multi-material image formation. The most common loss function is mean squared error with respect to supervised target images such as ground truth or higher-dose cases. However, we believe that the mean-squared error loss function has several issues for multi-material image formation. In this work, we present a new loss function which includes multiple noise realizations with separate weights on covariance and bias for joint denoising of all material bases. By modulating these weights, it is possible to tune the image quality of neural network output images. To demonstrate our proposed approach, we conducted a simulation of a water/calcium/gadolinium spectral CT imaging scenario using a deep neural network for multi-material image denoising. Our results show that by changing the weights of our proposed loss function, it is possible to control the tradeoff between variance and bias for individual materials as well as the control over the bias coupling between materials.

Local Linearity Analysis of Deep Learning CT Denoising Algorithms.

The rapid development of deep-learning methods in medical imaging has called for an analysis method suitable for non-linear and data-dependent algorithms. In this work, we investigate a local linearity analysis where a complex neural network can be represented as piecewise linear systems. We recognize that a large number of neural networks consists of alternating linear layers and rectified linear unit (ReLU) activations, and are therefore strictly piecewise linear. We investigated the extent of these locally linear regions by gradually adding perturbations to an operating point. For this work, we explored perturbations based on image features of interest, including lesion contrast, background, and additive noise. We then developed strategies to extend these strictly locally linear regions to include neighboring linear regions with similar gradients. Using these approximately linear regions, we applied singular value decomposition (SVD) analysis to each local linear system to investigate and explain the overall nonlinear and data-dependent behaviors of neural networks. The analysis was applied to an example CT denoising algorithm trained on thorax CT scans. We observed that the strictly local linear regions are highly sensitive to small signal perturbations. Over a range of lesion contrast from 0.007 to 0.04 mm-1, there is a total of 33992 linear regions. The Jacobians are also shift-variant. However, the Jacobians of neighboring linear regions are very similar. By combining linear regions with similar Jacobians, we narrowed down the number of approximately linear regions to four over lesion contrast from 0.001 to 0.08 mm-1. The SVD analysis to different linear regions revealed denoising behavior that is highly dependent on the background intensity. Analysis further identified greater amount of noise reduction in uniform regions compared to lesion edges. In summary, the local linearity analysis framework we proposed has the potential for us to better characterize and interpret the non-linear and data-dependent behaviors of neural networks.

Manifold Reconstruction of Differences: A Model-Based Iterative Statistical Estimation Algorithm With a Data-Driven Prior.

Manifold learning using deep neural networks been shown to be an effective tool for building sophisticated prior image models that can be applied to noise reduction in low-close CT. A manifold is a low-dimensional space that captures most of the variation in a class of data (e.g. chest CT images). This nonlinear reduction in dimensionality is one way to build up a sophisticated model of the features and structures present in image data. This nonlinear dimensionality reduction tends to learn features associated with the signal rather than the noise when provided with a large training dataset. However, the internal workings operation are typically not easy to understand. The results can be highly nonlinear and object-dependent. After training the model is fixed and during reconstruction is typically no check for fitness between the image estimates and the measured data. We propose a new iterative CT reconstruction algorithm, called Manifold Reconstruction of Differences (MRoD), which combines physical and statistical models with a data-driven prior based on manifold learning. The MRoD algorithm involves estimating a manifold component, approximating common features among all patients, and the difference component which has the freedom to fit the measured data. By applying a sparsity-promoting penalty to the difference image rather than a hard constraint to the manifold, the MRoD algorithm is able to reconstruct features which are not present in the training data. The difference component itself may be independently useful. While the manifold captures typical patient features (e.g. healthy anatomy), the difference image highlights patient-specific elements (e.g. pathology). In this work, we present a description of an optimization framework which combines trained manifold-based modules with physical modules. We present a simulation study using anthropomorphic lung data showing that the MRoD algorithm can both isolate differences between a particular patient and the typical distribution, but also provide significant noise reduction with less bias than a typical penalized likelihood estimator in composite manifold plus difference reconstructions.

A prototype spatial-spectral CT system for material decomposition with energy-integrating detectors.

PURPOSE: Spectral CT has great potential for a variety of clinical applications due to improved tissue and material discrimination over conventional single-energy CT. Many clinical and preclinical spectral CT systems have two spectral channels enabling dual-energy CT. Strategies include split filtration, dual-layer detectors, photon-counting detectors, and kVp switching. The motivation for this work is the development of an x-ray source spectral modulation device with three or more spectral channels to enable high-sensitivity multi-material decomposition with energy-integrating detectors. MATERIALS AND METHODS: We present spatial-spectral filters which are a new x-ray source modulation technology with the potential for additional channel diversity. The filtering device consists of an array of K-edge materials which divide the x-ray beam into spectrally varied beamlets. This design allows for an arbitrary number of spectral channels-trading off spatial and spectral information. We use a one-step model-based material decomposition (MBMD) algorithm to iteratively estimate material density images directly from the spatial-spectral CT data. In this work, we present a prototype spatial-spectral filter integrated with an x-ray CT test bench. The filter is composed of an array of tin, erbium, tantalum, and lead filter tiles which spatially modulate the system spectral sensitivity pattern. In a simulation study, we investigate the particular problem of mis-calibration between the data acquisition and the reconstruction model. With an understanding of the required model accuracy, we present a spectral calibration method to estimate the critical model parameters. To demonstrate feasibility of the spatial-spectral filter with a calibrated beamlet model, we collected a spatial-spectral CT scan of a multicontrast-enhanced phantom containing water, iodine, and gadolinium solutions. RESULTS: With simulations, we show that material decomposition is possible with spatial-spectral-filtered CT data, and we demonstrate the importance of a well-calibrated physical model. We find a 50% increase in error for focal spot model mismatch of 0.27mm and gap width model mismatch of 16 mµ. With physical results, we demonstrate that the calibrated system model is in close agreement with the measured data, and that the reconstructed material density images match the ground truth concentrations for the multicontrast phantom. Empirical results indicate gadolinium density estimation had an error of 17-58% mostly due to a systematic constant bias of 0.30-0.60 mg/ml. Water density estimates are within 1% and iodine estimates are within 10% of ground truth. CONCLUSION: These preliminary results demonstrate the potential of spatial-spectral filters to enable multicontrast imaging. Moreover, this device is compatible with energy-integrating detectors and so provides a feasible modification to enable spectral CT imaging with existing single-energy systems.

Spectral CT using a fine grid structure and varying x-ray incidence angle.

PURPOSE: Interest in spectral computed tomography (CT) for diagnostics and therapy evaluation has been growing. Data acquisitions with distinct spectral sensitivities provide the ability to discriminate multiple materials, quantitative density estimates, and reduced artifacts due to energy dependencies. We introduce a novel spectral CT concept that includes a fine-pitch grid structure for prefiltration of the x-ray beam. METHODS: We develop physical models for grid designs and illustrate the basic operating principles wherein small angulations of the incident x rays results significant filtration and spectral shaping of the beam. We fabricate a prototype grid with tungsten lamellae. We compare x-ray spectra induced by this filter as a function of incidence angle in both simulation students and in physical measurements. The grid is also integrated onto a CT test bench where we scanned an iodinated phantom with clinically relevant concentrations (5, 10, 20, and 50 mgI/mL) to demonstrate the ability to perform spectral CT acquisitions and material decomposition. RESULTS: X-ray spectrometer measurements reveal diverse and controllable spectral shaping with small angle changes that are in agreement with simulation studies. Critical angles where the characteristics of the induced spectrum changes dramatically are identified. Reconstructions of projection data for two angulations separated by 2° was reconstructed and material decomposition into iodine and water images shows good agreement with the known iodine concentrations. CONCLUSIONS: This work demonstrates the feasibility of the grid-based approach to enable spectral CT data acquisitions and accurate material decompositions. On-going and future studies will investigate the potential of this novel concept as a relatively simple upgrade to standard energy-integrating CT.

High-resolution model-based material decomposition in dual-layer flat-panel CBCT.

PURPOSE: Spectral CT uses energy-dependent measurements that enable material discrimination in addition to reconstruction of structural information. Flat-panel detectors (FPDs) have been widely used in dedicated and interventional systems to deliver high spatial resolution, volumetric cone-beam CT (CBCT) in compact and OR-friendly designs. In this work, we derive a model-based method that facilitates high-resolution material decomposition in a spectral CBCT system equipped with a prototype dual-layer FPD. Through high-fidelity modeling of multilayer detector, we seek to avoid resolution loss that is present in more traditional processing and decomposition approaches. METHOD: A physical model for spectral measurements in dual-layer flat-panel CBCT is developed including layer-dependent differences in system geometry, spectral sensitivities, and detector blur (e.g., due to varied scintillator thicknesses). This forward model is integrated into a model-based material decomposition (MBMD) method based on minimization of a penalized weighted least-squared (PWLS) objective function. The noise and resolution performance of this approach was compared with traditional projection-domain decomposition (PDD) and image-domain decomposition (IDD) approaches as well as one-step MBMD with lower-fidelity models that use approximated geometry, projection interpolation, or an idealized system geometry without system blur model. Physical studies using high-resolution three-dimensional (3D)-printed water-iodine phantoms were conducted to demonstrate the high-resolution imaging performance of the compared decomposition methods in iodine basis images and synthetic monoenergetic images. RESULTS: Physical experiments demonstrate that the MBMD methods incorporating an accurate geometry model can yield higher spatial resolution iodine basis images and synthetic monoenergetic images than PDD and IDD results at the same noise level. MBMD with blur modeling can further improve the spatial-resolution compared with the decomposition results obtained with IDD, PDD, and MBMD methods with lower-fidelity models. Using the MBMD without or with blur model can increase the absolute modulation at 1.75 lp/mm by 10% and 22% compared with IDD at the same noise level. CONCLUSION: The proposed model-based material decomposition method for a dual-layer flat-panel CBCT system has demonstrated an ability to extend high-resolution performance through sophisticated detector modeling including the layer-dependent blur. The proposed work has the potential to not only facilitate high-resolution spectral CT in interventional and dedicated CBCT systems, but may also provide the opportunity to evaluate different flat-panel design trade-offs including multilayer FPDs with mismatched geometries, scintillator thicknesses, and spectral sensitivities.