3D deep encoder-decoder network for fluorescence molecular tomography.

Fluorescence molecular tomography (FMT) is a promising and noninvasive in vivo functional imaging modality. However, the quality of FMT reconstruction is limited by the simplified linear model of photon propagation. Here, an end-to-end three-dimensional deep encoder-decoder (3D-En-Decoder) network is proposed to improve the quality of FMT reconstruction. It directly establishes the nonlinear mapping relationship between the inside fluorescent source distribution and the boundary fluorescent signal distribution. Thus the reconstruction inaccuracy caused by the simplified linear model can be fundamentally avoided by the proposed network. Both numerical simulations and phantom experiments were carried out, and the results demonstrated that the 3D-En-Decoder network can greatly improve image quality and significantly reduce reconstruction time compared with conventional methods.

End-to-end deep neural network for optical inversion in quantitative photoacoustic imaging.

An end-to-end deep neural network, ResU-net, is developed for quantitative photoacoustic imaging. A residual learning framework is used to facilitate optimization and to gain better accuracy from considerably increased network depth. The contracting and expanding paths enable ResU-net to extract comprehensive context information from multispectral initial pressure images and, subsequently, to infer a quantitative image of chromophore concentration or oxygen saturation (sO2). According to our numerical experiments, the estimations of sO2 and indocyanine green concentration are accurate and robust against variations in both optical property and object geometry. An extremely short reconstruction time of 22 ms is achieved.

Direct reconstruction method for time-domain fluorescence molecular lifetime tomography.

For the reconstruction of time-domain fluorescence molecular lifetime tomography, conventional methods based on the Laplace or Fourier transform utilize only part of the information from the measurement data, and rely on the selection of transformation factors. To make the best of all the measurement data, a direct reconstruction algorithm is proposed. The fluorescence yield map is first reconstructed with a full-time gate, and then an objective function for the inverse lifetime tomography (instead of the lifetime) is developed so as to avoid dealing with the singularity of the zero points in the lifetime image. Through simulations and physical phantom experiments, the proposed algorithm is demonstrated to have high localization accuracy for fluorescent targets, high quantification accuracy for fluorescence lifetime, and good contrast between different fluorescence targets.

Multi-segmented feature coupling for jointly reconstructing initial pressure and speed of sound in photoacoustic computed tomography.

SIGNIFICANCE: Photoacoustic computed tomography (PACT) is a fast-growing imaging modality. In PACT, the image quality is degraded due to the unknown distribution of the speed of sound (SoS). Emerging initial pressure (IP) and SoS joint-reconstruction methods promise reduced artifacts in PACT. However, previous joint-reconstruction methods have some deficiencies. A more effective method has promising prospects in preclinical applications. AIM: We propose a multi-segmented feature coupling (MSFC) method for SoS-IP joint reconstruction in PACT. APPROACH: In the proposed method, the ultrasound detectors were divided into multiple sub-arrays with each sub-array and its opposite counterpart considered to be a pair. The delay and sum algorithm was then used to reconstruct two images based on a subarray pair and estimated a direction-specific SoS, based on image correlation and the orientation of the subarrays. Once the data generated by all pairs of subarrays were processed, an image that was optimized in terms of minimal feature splitting in all directions was generated. Further, based on the direction-specific SoS, a model-based method was used to directly reconstruct the SoS distribution. RESULTS: Both phantom and animal experiments demonstrated feasibility and showed promising results compared with conventional methods, with less splitting and blurring and fewer distortions. CONCLUSIONS: The developed MSFC method shows promising results for both IP and SoS reconstruction. The MSFC method will help to optimize the image quality of PACT in clinical applications.

Depth-recognizable time-domain fluorescence molecular tomography in reflective geometry.

Conventional fluorescence molecular tomography (FMT) reconstruction requires photons penetrating the whole object, which limits its applications to small animals. However, by utilizing reflective photons, fluorescence distribution near the surface could be reconstructed regardless of the object size, which may extend the applications of FMT to surgical navigation and so on. Therefore, time-domain reflective fluorescence molecular tomography (TD-rFMT) is proposed in this paper. The system excites and detects the emission light from the same angle within a field of view of 5 cm. Because the detected intensities of targets depend strongly on the depth, the reconstruction of targets in deep regions would be evidently affected. Therefore, a fluorescence yield reconstruction method with depth regularization and a weighted separation reconstruction strategy for lifetime are developed to enhance the performance for deep targets. Through simulations and phantom experiments, TD-rFMT is proved capable of reconstructing fluorescence distribution within a 2.5-cm depth with accurate reconstructed yield, lifetime, and target position(s).

Feature coupling photoacoustic computed tomography for joint reconstruction of initial pressure and sound speed in vivo.

Photoacoustic imaging relies on diffused photons for optical contrast and diffracted ultrasound for high resolution. As a tomographic imaging modality, often an inverse problem of acoustic diffraction needs to be solved to reconstruct a photoacoustic image. The inverse problem is complicated by the fact that the acoustic properties, including the speed of sound distribution, in the image field of view are unknown. During reconstruction, subtle changes of the speed of sound in the acoustic ray path may accumulate and give rise to noticeable blurring in the image. Thus, in addition to the ultrasound detection bandwidth, inaccurate acoustic modeling, especially the unawareness of the speed of sound, defines the image resolution and influences image quantification. Here, we proposed a method termed feature coupling to jointly reconstruct the speed of sound distribution and a photoacoustic image with improved sharpness, at no additional hardware cost. Simulations, phantom studies, and in vivo experiments demonstrated the effectiveness and reliability of our method.

Streak artifact suppression in photoacoustic computed tomography using adaptive back projection.

For photoacoustic computed tomography (PACT), an insufficient number of ultrasound detectors can cause serious streak-type artifacts. These artifacts get overlaid on top of image features, and thus locally jeopardize image quality and resolution. Here, a reconstruction algorithm, termed Contamination-Tracing Back-Projection (CTBP), is proposed for the mitigation of streak-type artifacts. During reconstruction, CTBP adaptively adjusts the back-projection weight, whose value is determined by the likelihood of contamination, to minimize the negative influences of strong absorbers. An iterative solution of the eikonal equation is implemented to accurately trace the time of flight of different pixels. Numerical, phantom and in vivo experiments demonstrate that CTBP can dramatically suppress streak artifacts in PACT and improve image quality.

Reconstruction of high-resolution early-photon tomography based on the first derivative of temporal point spread function.

For fluorescence molecular tomography, higher spatial resolution can be achieved using minimally scattered early photons. Conventional reconstruction methods of early photon tomography (EPT) are based on the integral of temporal point spread function (TPSF), which may lead to poor image quality due to systematic noise and time mismatch between the TPSF data and forward model. The derivative of the rising portion of TPSF is proposed to be used in EPT to increase the performance of reconstruction because the derivative is less sensitive to noise and time mismatch than the integral. A method based on projected Tikhonov regularization with the reconstructed result of steepest descent algorithm as a priori information is developed. Using the derivative of TPSF, the method can achieve high spatial resolution in phantom experiments and is capable of separating targets with an edge-edge distance of 1.5 mm.

Effects of temperature on multiparametric evaluation of hindlimb ischemia with dynamic fluorescence imaging.

Quantitative evaluation of hindlimb ischemia is essential for early diagnosis and therapy of peripheral arterial disease (PAD). Dynamic imaging using near-infrared (NIR) fluorophore indocyanine green (ICG) is a noninvasive and effective tool to monitor multiple vascular parameters including perfusion rate (PR), perfusion vascular density (PVD) and hemodynamics. It has been previously demonstrated that temperature changes could lead to significant variations of blood flow rate and vascular perfusion. In this paper, multiparametric evaluation of hindlimb ischemia was performed at different temperatures. Five different parameters were extracted from dynamic fluorescence imaging, including PR, PVD, rising time (Trise ), blood flow index (BFI) and mean fluorescence intensity (MFI). Temperatures varied from 15 °C to 40 °C were set on a mouse model of hindlimb ischemia. The aforementioned five parameters were obtained at each temperature. The results suggest that PVD, BFI and MFI could be effective indicators to distinguish ischemic tissues from normal tissues in mouse hindlimb at different temperatures. In contrast, PR is effective only when the temperature is higher than 25 °C, while Trise is effective only when the temperature is lower than 35 °C. The parameters of PVD, BFI and MFI could provide quantitative and comprehensive evaluation for PAD at different temperatures. (A) Bright-field image of the normal (left) and ischemic (right) hindlimbs. (B-D) Parametric images of perfusion vascular density, blood flow index and mean fluorescence intensity.

Multiparametric evaluation of hindlimb ischemia using time-series indocyanine green fluorescence imaging.

Peripheral arterial disease (PAD) can further cause lower limb ischemia. Quantitative evaluation of the vascular perfusion in the ischemic limb contributes to diagnosis of PAD and preclinical development of new drug. In vivo time-series indocyanine green (ICG) fluorescence imaging can noninvasively monitor blood flow and has a deep tissue penetration. The perfusion rate estimated from the time-series ICG images is not enough for the evaluation of hindlimb ischemia. The information relevant to the vascular density is also important, because angiogenesis is an essential mechanism for post-ischemic recovery. In this paper, a multiparametric evaluation method is proposed for simultaneous estimation of multiple vascular perfusion parameters, including not only the perfusion rate but also the vascular perfusion density and the time-varying ICG concentration in veins. The target method is based on a mathematical model of ICG pharmacokinetics in the mouse hindlimb. The regression analysis performed on the time-series ICG images obtained from a dynamic reflectance fluorescence imaging system. The results demonstrate that the estimated multiple parameters are effective to quantitatively evaluate the vascular perfusion and distinguish hypo-perfused tissues from well-perfused tissues in the mouse hindlimb. The proposed multiparametric evaluation method could be useful for PAD diagnosis. The estimated perfusion rate and vascular perfusion density maps (left) and the time-varying ICG concentration in veins of the ankle region (right) of the normal and ischemic hindlimbs.

Facilitating in vivo tumor localization by principal component analysis based on dynamic fluorescence molecular imaging.

Fluorescence molecular imaging has been used to target tumors in mice with xenograft tumors. However, tumor imaging is largely distorted by the aggregation of fluorescent probes in the liver. A principal component analysis (PCA)-based strategy was applied on the in vivo dynamic fluorescence imaging results of three mice with xenograft tumors to facilitate tumor imaging, with the help of a tumor-specific fluorescent probe. Tumor-relevant features were extracted from the original images by PCA and represented by the principal component (PC) maps. The second principal component (PC2) map represented the tumor-related features, and the first principal component (PC1) map retained the original pharmacokinetic profiles, especially of the liver. The distribution patterns of the PC2 map of the tumor-bearing mice were in good agreement with the actual tumor location. The tumor-to-liver ratio and contrast-to-noise ratio were significantly higher on the PC2 map than on the original images, thus distinguishing the tumor from its nearby fluorescence noise of liver. The results suggest that the PC2 map could serve as a bioimaging marker to facilitate in vivo tumor localization, and dynamic fluorescence molecular imaging with PCA could be a valuable tool for future studies of in vivo tumor metabolism and progression.

Self-guided reconstruction for time-domain fluorescence molecular lifetime tomography.

Fluorescence probes have distinct yields and lifetimes when located in different environments, which makes the reconstruction of fluorescence molecular lifetime tomography (FMLT) challenging. To enhance the reconstruction performance of time-domain (TD) FMLT with heterogeneous targets, a self-guided L 1 regularization projected steepest descent (SGL1PSD) algorithm is proposed. Different from other algorithms performed in time domain, SGL1PSD introduces a time-resolved strategy into fluorescence yield reconstruction. The algorithm consists of four steps. Step 1 reconstructs the initial yield map with full time gate strategy; steps 2­4 reconstruct the inverse lifetime map, the yield map, and the inverse lifetime map again with time-resolved strategy, respectively. The reconstruction result of each step is used as a priori for the reconstruction of the next step. Projected iterated Tikhonov regularization algorithm is adopted for the yield map reconstructions in steps 1 and 3 to provide a solution with iterative refinement and nonnegative constraint. The inverse lifetime map reconstructions in steps 2 and 4 are based on L 1 regularization projected steepest descent algorithm, which employ the L 1 regularization to reduce the ill-posedness of the high-dimensional nonlinear problem. Phantom experiments with heterogeneous targets at different edge-to-edge distances demonstrate that SG

Automatic selection of regularization parameters for dynamic fluorescence molecular tomography: a comparison of L-curve and U-curve methods.

Dynamic fluorescence molecular tomography (FMT) is a promising technique for the study of the metabolic process of fluorescent agents in the biological body in vivo, and the quality of the parametric images relies heavily on the accuracy of the reconstructed FMT images. In typical dynamic FMT implementations, the imaged object is continuously monitored for more than 50 minutes. During each minute, a set of the fluorescent measurements is acquired and the corresponding FMT image is reconstructed. It is difficult to manually set the regularization parameter in the reconstruction of each FMT image. In this paper, the parametric images obtained with the L-curve and U-curve methods are quantitatively evaluated through numerical simulations, phantom experiments and in vivo experiments. The results illustrate that the U-curve method obtains better accuracy, stronger robustness and higher noise-resistance in parametric imaging. Therefore, it is a promising approach to automatic selection of the regularization parameters for dynamic FMT.

Fast direct reconstruction strategy of dynamic fluorescence molecular tomography using graphics processing units.

Dynamic fluorescence molecular tomography (DFMT) is a valuable method to evaluate the metabolic process of contrast agents in different organs in vivo, and direct reconstruction methods can improve the temporal resolution of DFMT. However, challenges still remain due to the large time consumption of the direct reconstruction methods. An acceleration strategy using graphics processing units (GPU) is presented. The procedure of conjugate gradient optimization in the direct reconstruction method is programmed using the compute unified device architecture and then accelerated on GPU. Numerical simulations and in vivo experiments are performed to validate the feasibility of the strategy. The results demonstrate that, compared with the traditional method, the proposed strategy can reduce the time consumption by ∼90% without a degradation of quality.

Nonlinear greedy sparsity-constrained algorithm for direct reconstruction of fluorescence molecular lifetime tomography.

In order to improve the spatial resolution of time-domain (TD) fluorescence molecular lifetime tomography (FMLT), an accelerated nonlinear orthogonal matching pursuit (ANOMP) algorithm is proposed. As a kind of nonlinear greedy sparsity-constrained methods, ANOMP can find an approximate solution of L0 minimization problem. ANOMP consists of two parts, i.e., the outer iterations and the inner iterations. Each outer iteration selects multiple elements to expand the support set of the inverse lifetime based on the gradients of a mismatch error. The inner iterations obtain an intermediate estimate based on the support set estimated in the outer iterations. The stopping criterion for the outer iterations is based on the stability of the maximum reconstructed values and is robust for problems with targets at different edge-to-edge distances (EEDs). Phantom experiments with two fluorophores at different EEDs and in vivo mouse experiments demonstrate that ANOMP can provide high quantification accuracy, even if the EED is relatively small, and high resolution.