Parametric Image

About: Parametric Image is a research topic. Over the lifetime, 311 publications have been published within this topic receiving 6095 citations.

Omni-directional Image Generation Algorithm with Parametric Image Compensation

Abstract: This paper proposes an omni-directional image generation algorithm with parametric image compensation. The algorithm generates an omni-directional image by transforming each planar image to the spherical image on spherical coordinate. Parametric image compensation method is presented in order to compensate vignetting and illumination distortions caused by properties of a camera system and lighting condition. The proposed algorithm can generates realistic and seamless omni-directional video and synthesize any point of view from the stitched omni-directional image on the spherical image. Experimental results show that the proposed omni-directional system with vignetting and illumination compensation is approximately better than that which does not consider the said effects.

Direct Estimation of Input Function Based on Fine-tuned Deep Learning Method in Dynamic PET Imaging

Abstract: 1394 Objectives: Dynamic PET provides more accurate quantitative information and richer disease information than static PET. However, routine clinical application of dynamic PET imaging is limited by invasive arterial blood sampling (or manually annotated image-derived blood activity) for use as an input function. This study aimed to develop a deep learning-based method to directly estimate the input function from dynamic PET data without any manual assistance. Methods: Our study consists of two clinical datasets. The first group of data includes dynamic FDG scans performed on the United Imaging uMI 510 PET/CT system in the Chinese PLA General Hospital. 35 subjects (25 male, 10 females; 13 healthy; age 15-73y) were scanned. Image-derived blood activity was used as input function. The second dataset (n= 26 healthy subjects) includes 90-min dynamic brain data after bolus injection of 11C-DPA-713(DPA) performed on a Siemens HRRT PET system at Johns Hopkins University. All DPA PET data in this study were acquired from individuals with high affinity binding genotype for the 18 kDa translocator protein target and included input function data acquired through arterial blood sampling and radiometabolite measurements. There were 30 dynamic frames in each scan. The input function value for each dynamic frame was interpolated from the measured input function curve. Each dataset was split into 3 parts: 70% as training set, 20% as validation set and 10% as test set. Two deep learning networks, namely the raw model and the fine-tuned model were investigated. The raw model contains a down-sampling convolutional module to extract image features and a fully connected regression module to predict the input function. The 3-D dynamic image and reference image are stacked to form a two-channel 4-D input. For FDG tracer, the reference image refers to image reconstructed from data acquired in 0 to 20 minutes scan. For DPA, the mean image of all dynamic images served as the reference image. L1-loss is chosen as the loss function based on controlled experiment results. In the fine-tuned model, the 1-D vector acquired from the convolutional module are concatenated with medical information including patient’s age, weight, height, injected dose, frame time. Data augmentation methods such as random center crop & resize, rotation, translation and scaling were used in training fine-tuned model. The input function for DPA was normalized according to its integral value before reconstruction. Indirect voxel-based reconstruction methods were used: Patlak analysis with start time (t*) of 20 minutes for Ki (FDG) and Logan analysis with t* of 30 minutes for VT (DPA). To investigate the performance of the proposed method, the root mean square error (RMSE) was calculated between the parametric images reconstructed with true and predicted input function. Results: The predicted input function shows similarity in shape to the standard input function. The prediction error in early stage was larger than in later stage, especially for the peak values in early stage. The fine-tuned model enhances prediction accuracy in the early stage, while the improvement in later stage was not obvious. The visual quality of reconstructed Ki and VT parametric image of the fine-tuned model was apparently better than that of raw model. The RMSE of the fine-tuned model was much lower than the raw model. Conclusions: The proposed fine-tuned deep learning-based method was able to estimate the input function directly from dynamic images. The high accuracy of the reconstructed parametric image using the predicted input function supports pursuit of this method for clinical application. In future work we will investigate to further improve the accuracy of the predicted input function. Support: The research was supported by the National Natural Science Foundation of China (No. 81727807, No.11575096, No. 11605008) and National Key Research and Development (R&D) Plan of China (Grant ID. 2019YFF0302503 and 2016YFC0105405).

Parametric imaging of gene expression in biopsy samples utilizing the reverse transcriptase-polymerase chain reaction method

Abstract: A method is described for generating and showing the profile of one or more genes across a tissue sample. Tissue removed from the body is sliced in to thin sheets and those sheets are then divided into small portions each portion being identified as to a location in the sheet. The image of the thin slice and the position of each small portion thereof is recorded in a computer in a manner that can generate an image of the slice. Each small portion is subject to RT-PCR to identify the presence and quantity of one or more genes therein. The portion-specific data is then entered into the computer and an image of the slice is generated showing the gene specific characteristics of each small portion. The result is a parametric image of the entire slice which allows the visualization of the gene expression within each portion which can then be compared with other images of the same or adjacent tissue.

Sparsity Constrained Direct Parametric Reconstruction in Dynamic PET Myocardial Perfusion Imaging

Abstract: 110 Objectives: Dynamic myocardial perfusion imaging (MPI) with PET serves an important role in diagnosis and prognosis of patents with suspected or known coronary artery disease. Conventional myocardial blood flow (MBF) quantification reconstructs a series of dynamic frames and applies a kinetic model to the reconstructed image sequences for measuring the tracer uptake rate K1. This approach leads to noisy K1 estimation due to very limited counts in individual time frames. The goal of this study is to develop a sparsity regularized direct parametric reconstruction algorithm, which collectively uses data from all the dynamic frames while imposing a dictionary learning (DL) based sparsity constraint on K1 spatial variation. Methods: The direct parametric reconstruction is accomplished by relating parametric images to dynamic PET data through a nonlinear transform containing the one-tissue compartment model and the imaging system matrix. To suppress the undesirable noise propagation in this ill-posed inverse problem, we impose a sparsity regularization on the K1 image leading to a penalized log-likelihood function for maximization. The sparsity constraint is constructed as the difference between the estimated K1 image and its sparse representation based on the learned dictionary from a self-created hollow sphere. A two-stage iterative approach is adopted to solve this optimization problem. In stage one, we calculate the sparse representation of the current estimation of K1 given the learned dictionary. In stage two, the penalized log-likelihood function is optimized with the sparsity penalty term fixed. Applying optimization transfer, we construct separable surrogate functions for the log-likelihood term and the sparsity constraint term, respectively. The combined surrogate function, which is solvable by convenient voxel-wise optimization is maximized by the damped Newton method. To evaluate the proposed algorithm, we simulated two sets of realistic Rb-82 dynamic MPI PET data, one with normal MBF (K1=1.48) and the other with regionally reduced MBF (K1=1.13). Using the XCAT phantom, PET image frames were created by assigning the activities integrated from the multiple organ time activity curves based on clinical measurement. We performed analytic simulations for the geometry of a GE RX PET scanner to generate 20 noise realizations for each dynamic dataset. By assessing the ensemble noise versus bias tradeoff of the normal and abnormal K1 on the region of interest, we compared the proposed method, the conventional method with and without post filtering, and a quadratic penalty regularized direct parametric reconstruction method. We also evaluated the tradeoff of the ensemble noise and the contrast between the normal and the defect K1 for abnormal MBF detectability. Results: For the regional normal K1 estimation, the mean and the ensemble normalized standard deviation (EnNSD) across 20 noise realizations obtained by the conventional method without and with post filtering, the quadratic penalty regularized direct algorithm, and the proposed method are 1.43+/-0.45, 1.10+/-0.21, 1.36+/-0.22, and 1.45+/-0.22. For the abnormal case, the corresponding regional mean+/-EnNSD are 1.13+/-0.64, 0.78+/-0.26, 1.02+/-0.27, and 1.07+/-0.27, respectively. In both cases, post filtering in the conventional method reduces noise at the cost of introducing large bias. The proposed sparsity constraint outperforms the quadratic penalty, achieving similar noise but reduced bias resulting in better recovered contrast. Conclusions: We developed a sparsity constrained direct parametric image reconstruction algorithm that incorporates the DL based regularization on the K1 parametric image. Using simulated dynamic MPI PET data, we demonstrated its better performance in K1 estimation and abnormal K1 detection compared with conventional methods. The proposed method shows its potential to advance MBF quantification in dynamic PET MPI.

Parametric Image Generation with Neural Networks

Abstract: An artificial neural network (ANN) is a trainable algorithm that can learn to produce an output appropriate for a given input. Such networks can be applied in a wide variety of pattern recognition tasks, including parameter estimation. The major advantages of using ANNs for parameter estimation are speed and noise tolerance. A two-layer network was used to determine the rate constants and metabolic rate (MRfdg) from dynamic PET images acquired after injection of [18F]fluorodeoxyglucose (FDG). The number of input points was 24, which represented a well-sampled tissue time-activity-curve (TAC) over 60 min after injection of FDG. One thousand noisy training data sets were generated, using the arterial plasma TAC, for each subject (computer time: ≈ 1 min). The network was trained with the backpropagation algorithm using the 1000 data sets (time: ≈ 15 min). Images were generated using the weights determined by training (time: ≈ 7 sec). Times are for a Macintosh 7100/80. The MRfdg images were of high quality (better contrast than the integrated images and less noisy than Patlak images). Parametric images (K1, k2, or k3) were quite noisy. Training the ANN with an average plasma TAC instead of the individual TACs resulted in MRfdg images of equal quality. Preliminary data from four subjects suggest it may be possible to generate quantitatively accurate images by scaling with the count from a single blood sample taken 30 min after injection.