Imaging diagnostic method for diagnosing cardiovascular disease

Abstract

An image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, comprising: (a) image preprocessing steps including: (i) reconstructing dynamic cine 3D tomographic myocardial perfusion imaging (MPI) data; (ii) denoising; (iii) extracting a blood input function from a region of interest (ROI) of the left ventricular blood chamber; (iv) estimating the distribution volume (DV) obtained from the ratio of uptake rate to washout rate (K1 / k2) to improve estimation of K1 and total blood volume (TBV) and myocardial blood flow measurement; and (v) normalizing the data by dividing by the maximum value of the blood input function; (b) evaluating the preprocessed signals in (a) to generate K1 and TBV parametric maps using an artificial neural network; and (c) postprocessing the K1, k2, and TBV parametric maps, as well as resting and stress myocardial blood flow.

Description

[Technical Field]

[0001] The present invention relates to nuclear imaging and medical care for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve. [Background technology]

[0002] Nuclear medicine uses radioactive materials for treatment and diagnostic imaging. There are various types of imaging procedures that utilize a radiopharmaceutical dose. A radiopharmaceutical dose can be injected into a patient before or during an imaging procedure. The radiopharmaceutical is absorbed by or attaches to the cells of the patient's target organ, emitting radiation. A scanner or detector in the imaging process detects the emitted radiation and produces an image of the organ. For example, to image body tissues such as the heart muscle, a patient is injected with Rb-82 (rubidium-82). The imaging procedure detects the Rb-82 radiation to obtain clearer images of the heart muscle and diagnose any associated problems.

[0003] Radioisotopes play an important role in the diagnosis and mitigation of various medical conditions. Radioisotopes are used, for example, in the treatment of cancer. 60 Co, in the treatment of hyperthyroidism 131 I. Breath testing 14 C. As a tracer in myocardial perfusion imaging 99m Tc and 82 Rb, etc.

[0004] Additionally, rubidium-82 is produced in situ by the radioactive decay of strontium-82. Rubidium elution systems utilize the dose of rubidium-82 produced by elution in a radioisotope generator and inject the radioactive solution into the patient.

[0005] Previous preclinical studies in dogs have shown that myocardial uptake of Rb-82 radionuclides is directly related to myocardial blood flow (MBF). Dynamic cine myocardial perfusion imaging (MPI) using radioisotopes can accurately predict myocardial blood flow (MBF) and myocardial flow reserve (MFR). MBF estimation typically begins with segmentation of the left ventricular (LV) myocardium, followed by tracer kinetic modeling in a limited number of two-dimensional (2D) polar maps or segments. However, 2D polar map segments have drawbacks. 2D polar maps often fail to accurately depict cardiac disease associated with small regional perfusion defects. Therefore, an alternative method for myocardial perfusion imaging is to visualize three-dimensional (3D) parametric maps of MBF. However, visualizing 3D parametric maps of MBF has several drawbacks. Furthermore, small regional perfusion defects are not always easily visualized. Furthermore, generating three-dimensional (3D) parametric maps is time-consuming and unstable. These drawbacks discourage medical professionals from adopting 3D parametric maps of MBF for estimating myocardial blood flow. Therefore, an alternative approach is needed to generate more stable three-dimensional (3D) parametric maps in a shorter time frame to estimate myocardial blood flow and flow reserve.

[0006] Due to limitations of early-generation tracer delivery systems, Rb-82 PET imaging was performed using a single fixed dose for all patients. The undesirable effects of these older PET imaging systems can be mitigated to some extent by using advanced, latest-generation Rb-82 elution systems. We found that 3D parametric imaging of myocardial blood flow using Rb-82 PET can accurately estimate and / or predict myocardial blood flow (MBF) and / or myocardial flow reserve (MFR). Summary of the Invention

[0007] The present invention aims to provide an image processing method for quantitatively assessing myocardial blood flow (MBF) and / or myocardial flow reserve (MFR).

[0008] The objective of the present invention is to provide an alternative method for estimating and / or predicting quantitative myocardial blood flow (MBF) and / or myocardial flow reserve (MFR) using 3D voxel-wise parametric image data, which allows for better highlighting of small regional perfusion defects. More precisely, the inventors generate voxel-wise parametric maps to estimate myocardial blood flow (MBF) and / or myocardial flow reserve (MFR), fit the image series to a single tissue compartment model, and compare the projected data with a two-dimensional (2D) polar map of the left ventricle (LV). Thus, the present invention has the advantage of better highlighting small regional perfusion defects and generating regional perfusion and reserve values ​​that are independent of segmentation of the LV polar map.

[0009] The purpose of the present invention is to provide an image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve. To improve image quality, an image reconstruction algorithm has been developed, which uses AI algorithms to enhance the quality of image reconstruction and speed up image processing during myocardial perfusion imaging (MPI), thereby aiming to reduce nuclear medicine dose by up to 10 times.

[0010] The objective of the present invention is to generate blood flow parametric maps in an AI model with high accuracy and in a time frame acceptable for clinical use, enabling future clinical implementation.

[0011] It is an object of the present invention to provide an image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve, in which the 3D parametric images of MBF generated by the present invention also recommend calcium scoring.

[0012] One embodiment of the present invention provides an image processing method for quantitatively evaluating myocardial blood flow and myocardial blood flow reserve, the image processing method including the following steps a to c. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion imaging (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting a blood input function from the left ventricle blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) Estimating the volume of distribution (DV) obtained by the ratio of uptake rate to washout rate (K1 / k2) to stabilize and improve the estimation of K1 and total blood volume (TBV) and subsequent myocardial blood flow measurements; and (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Evaluating the individual signals preprocessed in step (a) to generate K1 and TBV parametric maps using an artificial neural network. c. Post-processing the K1 and TBV parametric maps and resting and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR) maps.

[0013] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, in which the image reconstruction of the array is a dynamic cine series of 3D tomographic voxels (i, j, k) obtained by PET reconstruction arranged over a number of time steps ti (i is 1 to N).

[0014] Further, one embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, wherein input signals are fed into a multi-layer perceptron (MLP), an artificial neural network and / or a convolutional neural network (CNN) and / or a long-short-term memory (LSTM) network to simultaneously predict uptake rate (K1), washout rate (k2) and total blood volume (TBV).

[0015] Other embodiments of the present invention include imaging methods for assessing quantitative myocardial perfusion and / or myocardial perfusion reserve, wherein an imaging agent or radionuclide is administered by an automated production and injection system and / or intravenous administration of radiopharmaceuticals produced by nuclear fission, neutron activation, cyclotron, generator and / or combinations thereof.

[0016] Yet another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, the method comprising the following steps a to d. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting the blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) Estimating the volume of distribution (DV) given by the ratio of uptake and washout rates (K1 / k2) to stabilize and improve the estimation of K1, k2 and total blood volume (TBV) and subsequent myocardial blood flow measurements; (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Applying the time series and blood input function at voxel (i, j, k) simultaneously to an artificial intelligence network to predict intakes K1, k2, and TBV. c. Post-processing the K1, k2, TBV parametric maps, where the post-processing includes: (i) and (ii) the following: (i) partial volume correction, and (ii) Extraction fractions to estimate myocardial blood flow (MBF) at rest and during stress. d. Post-processing the resting and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR), wherein the artificial neural network is selected from the group consisting of multi-layer perceptrons (MLPs), artificial neural networks (ANNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long-short-term memory recurrent neural networks (LSTM-RNNs), gated recurrent unit (GRU) networks, generative adversarial networks (GANs), deep machine learning, and / or combinations thereof. [Brief explanation of the drawings]

[0017] Further features and advantages of the present invention will be apparent from the accompanying drawings and from the detailed description that follows.

[0018] [Figure 1] FIG. 1 is a flow diagram in which the model takes a blood input function and a time series at a voxel and outputs K1, k2 and TBV (total blood volume). [Figure 2] FIG. 1 is a flow diagram in which the model takes a blood input function, a time series in voxels, a volume of distribution (DV), and outputs K1 and TBV (total blood volume). [Figure 3] FIG. 1 is a flow diagram of model training. [Figure 4] A comparison of ground truth and AI-generated parametric maps. DETAILED DESCRIPTION OF THE INVENTION

[0019] Currently, there is a need for improved radioisotope imaging procedures for estimating and / or predicting small regional dysfunctions or impairments related to myocardial blood flow (MBF) and / or myocardial flow reserve (MFR), in which image series are fitted to a tissue compartment model to generate voxel-wise parametric maps that compare the data projected onto two-dimensional (2D) polar maps of the target organ, such as the left ventricular (LV) myocardium. The inventors have found that generating regional flow and reserve values ​​can more clearly highlight small regional flow defects and is advantageous in that it does not rely on segmentation of the LV polar map. The inventors have found that alternative 3D parametric imaging methods of myocardial blood flow using radioisotopes can accurately estimate and / or predict myocardial blood flow (MBF) and / or myocardial flow reserve (MFR). The present invention can be more readily understood from the detailed description and embodiments of the invention below.

[0020] As used herein, the term "imaging" refers to techniques and processes used to create images of various parts of the human body for diagnostic and therapeutic purposes in digital health. These include radiography, X-ray fluorescence, magnetic resonance imaging (MRI), computed tomography (CT), medical ultrasound or endoscopic ultrasound, elastography, tactile imaging, thermography, medical photography, and nuclear medicine functional imaging techniques (such as positron emission tomography (PET), dynamic positron emission tomography, or single-photon emission computed tomography (SPECT)). Diagnostic imaging reveals the internal structure of the body and allows for the diagnosis and treatment of disease.

[0021] As used herein, the term "Positron Emission Tomography (PET)" refers to a functional imaging technique that uses radioactive substances known as radiotracers or radionuclides to visualize and measure changes in metabolic processes and physiological activities, including blood flow, local chemical composition, and absorption. Radionuclide isotopes commonly used in PET imaging include Rb-82 (rubidium-82), WaterO-15 (oxygen-15), F-18 (fluorine-18), Ga-68 (gallium-68), Cu-61 (copper-61), C-11 (carbon-11), N-13 (ammonia-13), Co-55 (cobalt-55), Zr-89 (zirconium-89), Cu-62, Cu-64, I-124, Tc-99m (technetium), Tl-201 (thallium-201), and FDG (fluorodeoxyglucose). The preferred radionuclide is Rb-82, which has a half-life of 75 seconds.

[0022] As used herein, the term "SPECT" refers to single-photon emission computed tomography. Single-photon emission computed tomography is a nuclear medicine tomography technique that uses gamma rays to provide true 3D information. This information is typically displayed as cross-sectional slices of the patient, but can be freely reformatted or manipulated as needed. The technique requires the patient to receive a gamma-ray-emitting radioisotope (nuclide), usually injected into the bloodstream. The marker radioisotope is typically bound to a specific ligand, resulting in a radioligand that binds to specific types of tissue. This delivers the ligand-radiopharmaceutical combination to the site of interest in the body, where it binds and the ligand concentration is assessed by a gamma camera. SPECT agents include: 99m Tc Technetium-99m( 99m Tc)-sestamibi, 99m These include Tc-tetrofosmin, In-111, Ga-67, Ga-68, and Tl-201 (thallium-201).

[0023] As used herein, the term "diagnosis" refers to the process of identifying a disease, condition, or injury from its signs and symptoms. Health history, physical examination, and tests such as blood tests, imaging, scans, and biopsies are used in diagnosis.

[0024] As used herein, the term "assessment" refers to the qualitative and / or quantitative evaluation of blood perfusion in a body part or region of interest (ROI).

[0025] The term "stress agent," as used herein, refers to an agent used to induce stress in a patient or subject during an imaging procedure. Stress agents according to the present invention include vasodilators (e.g., adenosine, adenosine triphosphate and its mimetics), A2A adenosine receptor agonists (e.g., regadenoson or the adenosine reuptake inhibitor dipyridamole), other pharmacological agents that increase blood flow to the heart, such as catecholamines (e.g., dobutamine, acetylcholine, papaverine, ergovine, etc.), or other external stimuli that increase blood flow to the heart, such as cold pressure, mental stress, or physical exercise.

[0026] As used herein, the terms "automated infusion system" or "radionuclide generation" and / or "infusion system" or "Rb-82 elution system" refer to a system for generating and / or injecting a radionuclide or radiotracer and administering it to a subject. The automated infusion system consists of a radioisotope generator, a dose calibration device, a computer, a control device, a display device, a radioactivity detector, a cabinet, a cart, a waste bottle, sensors, a shielding assembly, an alarm or warning mechanism, tubing, a source vial, a diluent or eluent, and valves. The automated infusion system can be communicatively or electronically coupled to an imaging system.

[0027] As used herein, the term "dose" refers to the dose of a radionuclide required to perform imaging in a subject. The dose of a radionuclide administered to a subject is between 0.01 MBq and 10,000 MBq.

[0028] As used herein, the terms "coronary artery disease" or "cardiovascular disease" refer to diseases of the major blood vessels. Cholesterol-containing deposits (plaque) and inflammation in the coronary arteries are the cause of coronary artery disease. Coronary arteries supply blood, oxygen, and nutrients to the heart. Plaque buildup narrows these arteries, reducing blood flow to the heart. Ultimately, reduced blood flow causes chest pain (angina), shortness of breath, and other signs and symptoms of coronary artery disease. Significant blockage of the arteries can lead to a heart attack. Myocardial infarction can be diagnosed by imaging of the myocardium and / or myocardial blood flow (MBF) under resting or pharmacological stress conditions, which can assess regional myocardial blood flow.

[0029] As used herein, the term "myocardial blood flow (MBF)" can be defined as the volume of blood passing through tissue at a given rate. MFR is the ratio of MBF at maximum coronary vasodilation to MBF at rest and is therefore affected by both resting and stress blood flow. MFR represents the relative reserve of the coronary circulation.

[0030] As used herein, the term "radionuclide" or "radioisotope" refers to an unstable form of a chemical element that emits radiation as it breaks down and becomes stable. Radionuclides can occur naturally or can be produced in a laboratory. In medicine, they are used for imaging and treatment.

[0031] As used herein, "Sr / Rb elution system" or " 82Sr / 82 The term "Sr / Rb elution system" refers to an injection system for producing a solution containing Rb-82, measuring the radioactivity in that solution, and injecting that solution into a subject to perform various studies on the subject's area of ​​interest.

[0032] As used herein, the term "linage counts" refers to the number of radioisotope decays acquired per unit time by a PET scanner.

[0033] As used herein, the term "generator" or "radioisotope generator" refers to a hollow column within a radiation-shielded container. The column is filled with ion exchange resin, and a radioisotope is loaded onto the resin. Radionuclide generators according to the present invention include: 99 Mo / 99m Tc, 90 Sr / 90 Y, 82 Sr / 82 Rb, 188 W / 188 Re, 68 Ge / 68 Ga, 42 Ar / 42 K. 44 Ti / 44 Sc, 52 Fe / 52m Mn, 72 Se / 72 As, 83 Rb / 83m Kr, 103 Pd / 103m Rh, 109 Cd / 109m Ag, 113 Sn / 113m In, 118 Te / 118 Sb, 132 Te / 132 I, 137 Cs / 137m Ba, 140 Ba / 140 La, 134 Ce / 134 La, 144 Ce / 144 Pr, 140 Nd / 140 Pr, 166 Dy / 166 Ho, 167 Tm / 167m Er, 172 Hf / 172 Lu, 178 W / 178 Ta, 191 Os / 191m Ir, 194 Os / 194 Ir, 226 Ra / 222 Rn, 225 Ac / 213 Bi, and 64 Zn / 61 Cu.

[0034] As used herein, the term "eluant" refers to a liquid or fluid used to selectively elute daughter radioisotopes from a generator column.

[0035] As used herein, the term "eluate" refers to the radioactive eluate after obtaining the daughter radioisotope from the generator column.

[0036] As used herein, the term "controller" refers to a computer or part thereof that is programmed to perform certain calculations, execute instructions, and control various activities of the elution system, either based on user input or automatically.

[0037] As used herein, the term "activity detector" refers to a component used to determine the amount of radioactivity present in the effluent from a generator, for example, prior to administering the effluent to a patient.

[0038] As used herein, the term "convolutional neural network (CNN)" refers to a system similar to a feedforward neural system. It is a type of artificial neural network used in time series and image processing, specifically designed to process pixel data. In deep learning, convolutional neural networks (CNN / ConvNet) are a type of deep neural network that are most commonly applied to the analysis of time series as well as natural and medical images. A blood input function is extracted from a region of interest (ROI) either manually or automatically. The average signal within the ROI is extracted at each time point to create a blood input function for the 1D signal.

[0039] As used herein, the term "Multi-layer perceptron (MLP)" refers to an example of an artificial neural network that is widely used to solve a variety of problems, including pattern recognition and interpolation.

[0040] As used herein, the term "recurrent neural network (RNN)" refers to a special type of artificial neural network adapted to work on time-series data or data containing sequences. Typical feedforward neural networks operate only on data points that are independent of each other. RNN methods include long-short-term memory (LSTM) or gated recurrent unit (GRU) networks. RNNs can be combined with CNNs to form networks such as CNN-LSTM.

[0041] As used herein, the term "Gated Recurrent Unit (GRU)" refers to a type of recurrent neural network (RNN) that has a low memory footprint. It is part of a specific model of recurrent neural network that uses connections through a series of nodes to perform machine learning tasks related to memory and clustering. It includes a gating mechanism for recurrent neural networks.

[0042] As used herein, the term "voxel" refers to a value on a regular grid in three-dimensional (3D) computer graphics. A voxel is short for volume pixel, which is the smallest distinguishable cubic portion of a 3D image. Voxelization is the process of adding depth to an image using a series of cross-sectional images called a volume dataset. These cross-sectional (or slice) images are made up of pixels. Collectively, the pixels and slices form a three-dimensional (3D) division of the image space into volume elements (voxels), which form a 3D scalar field.

[0043] As used herein, the term "Tissue Response Function (TRF)" refers to tracer kinetic modeling, which is used to estimate physiological parameters such as myocardial blood flow (MBF) by mapping or transforming the shape of the "arterial input function (AIF)" to the shape of the TRF.

[0044] As used herein, the term "blood input function" is commonly known as the arterial input function (AIF) and is defined as the concentration of tracer in an artery measured over time by placing a region of interest.

[0045] One embodiment of the present invention is an image processing method for quantitatively evaluating myocardial blood flow and myocardial blood flow reserve, which includes the following steps a to c. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion imaging (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting the blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) Estimating the volume of distribution (DV) given by the ratio of uptake and washout rates (K1 / k2) to stabilize and improve the estimation of K1, k2 and total blood volume (TBV) and subsequent myocardial blood flow measurements; and (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Evaluating the individual signals preprocessed in step (a) to generate K1 and TBV parametric maps using an artificial neural network. c. K1, k2, ...K to estimate myocardial flow reserve (MFR) maps and / or coronary flow reserve (CFR) maps n , X 2 , R 2 Post-processing of , volume of distribution (DV), and TBV parametric maps, as well as resting and stress myocardial blood flow.

[0046] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, wherein the image reconstruction of the array is a dynamic series of PET-reconstructed 3D slice voxels (i, j, k) arranged over a number of time steps ti (i is 1 to N);

[0047] Another embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial flow reserve assessment, where regions of interest (ROI) can be set manually and / or automatically.

[0048] Furthermore, one embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial flow reserve assessment, where the data is normalized by dividing by the maximum value of the blood input function.

[0049] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, in which data of the blood input function is normalized to a value between 0 and 1.

[0050] Another embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, wherein input signals are fed into multi-layer perceptrons and / or artificial neural networks and / or generative adversarial networks (GANs) and / or convolutional neural networks (CNNs) and / or long short-term memory (LSTM) networks to simultaneously predict uptake fraction (K1), washout fraction (k2), distribution volume (DV), and total blood volume (TBV).

[0051] Additionally, one embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, generating regional blood flow and reserve values ​​from images to highlight small local blood flow defects.

[0052] One embodiment of the present invention includes an image processing method for assessing quantitative myocardial perfusion and myocardial perfusion reserve, wherein the artificial neural network is selected from the group consisting of a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network (CNN), a recurrent neural network (RNN), a long-short-term memory recurrent neural network (LSTM-RNN), a gated recurrent unit (GRU) network, deep machine learning, and / or combinations thereof.

[0053] Another embodiment of the present invention includes an image processing method for assessing quantitative myocardial blood flow and myocardial blood flow reserve, wherein an artificial neural network is input with multiple layers to estimate distribution volume (DV), the multiple layers being selected from the group consisting of initial layers, intermediate layers, final layers of the network, or a combination thereof.

[0054] One embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial flow reserve assessment, where a model predicts k2 (washout fraction) values.

[0055] Furthermore, one embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial flow reserve assessment, wherein anisotropic diffusion filtering is performed using a Gaussian filter.

[0056] One embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial flow reserve assessment, using a 1D signal CNN-LSTM performed voxel-wise to estimate K, k, and total blood volume (TBV), producing more accurate myocardial blood flow (MBF) estimates.

[0057] Other embodiments of the present invention include imaging methods for assessing quantitative myocardial blood flow and myocardial blood flow reserve, wherein images are characterized by administering Rb-82, O-15, N-13, F-18, Cu-62, Tc-99m, Tl-201, and / or combinations thereof.

[0058] One embodiment of the present invention includes an image processing method for assessing quantitative myocardial blood flow and myocardial flow reserve, where images are characterized by administering Rb-82 to highlight small regional blood flow defects during rest and stress PET perfusion imaging.

[0059] One embodiment of the present invention includes an imaging method for assessing quantitative myocardial perfusion and myocardial perfusion reserve, wherein an imaging agent or radionuclide is administered by an automated production and injection system and / or intravenous administration of radiopharmaceuticals produced by nuclear fission, neutron activation, cyclotron and / or generator.

[0060] One embodiment of the present invention includes an imaging method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, wherein the automated radioisotope generation and injection system includes a Rb-82 elution system.

[0061] Other embodiments of the present invention include image processing methods for quantitative myocardial perfusion and myocardial flow reserve assessment, wherein the image processing method includes fitting the acquired images to a single tissue compartment model or a multi-tissue compartment model to determine myocardial perfusion reserve and / or coronary flow reserve by predicting a ratio of myocardial perfusion stress to rest, and evaluating the acquired images using a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network (CNN), a recurrent neural network (RNN), a long-short-term memory recurrent neural network (LSTM-RNN), a gated recurrent unit (GRU) network, deep machine learning, a deep neural network, an artificial neural network, and / or combinations thereof to diagnose a disease state.

[0062] Furthermore, one embodiment of the present invention includes an imaging method for assessing quantitative myocardial blood flow and myocardial blood flow reserve, wherein the imaging includes positron emission tomography (PET), dynamic positron emission tomography, single photon emission computed tomography (SPECT), magnetic resonance imaging (MRI), computed tomography (CT), and / or combinations thereof.

[0063] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, the method comprising the following steps a to c. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting the blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest; (v) Estimating the volume of distribution (DV) given by the ratio of uptake and washout rates (K1 / k2) to stabilize and improve the estimation of K1, k2 and total blood volume (TBV) and subsequent myocardial blood flow measurements; (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Applying the time series and blood input function at voxel (i, j, k) simultaneously to an artificial intelligence network to predict intakes K1, k2, and TBV. c. Post-processing the K1, k2, TBV parametric maps, where the post-processing includes: (i) and (ii) the following: (i) partial volume correction, and (ii) Extraction fractions to estimate myocardial blood flow (MBF) at rest and during stress. d. Post-processing rest and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR).

[0064] Yet another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, the method comprising the following steps a to d. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting blood input functions from left ventricular blood regions of interest (ROIs) and other arterial blood regions of interest; (v) Estimating the volume of distribution (DV) given by the ratio of uptake and washout rates (K1 / k2) to stabilize and improve the estimation of K1, k2 and total blood volume (TBV) and subsequent myocardial blood flow measurements; (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Applying the time series and blood input function at voxel (i, j, k) simultaneously to an artificial intelligence network to predict intakes K1, k2 and TBV. c. Post-processing the K1, k2, TBV parametric maps, where the post-processing includes: (i) and (ii) the following: (i) partial volume correction, and (ii) Extraction fractions to estimate myocardial blood flow (MBF) at rest and during stress. d. Post-processing the resting and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR), wherein the artificial neural network is selected from the group consisting of multi-layer perceptrons (MLPs), artificial neural networks (ANNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long-short-term memory recurrent neural networks (LSTM-RNNs), gated recurrent unit (GRU) networks, generative adversarial networks (GANs), deep machine learning, and / or combinations thereof.

[0065] Yet another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, the method comprising the following steps a to d. a. Preprocessing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) Isolating the value of voxel (i,j,k) at each time point ti (i = 1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting the blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) Estimating the volume of distribution (DV) given by the ratio of uptake and washout rates (K1 / k2) to stabilize and improve the estimation of K1, k2 and total blood volume (TBV) and subsequent myocardial blood flow measurements; (vi) normalizing the data by dividing by the maximum value of the blood input function. b. Applying the time series and blood input function at voxel (i, j, k) simultaneously to an artificial intelligence network to predict intake K1 and TBV, where the average R 2 The value is 0.9 to 1. c. Post-processing the K1 and TBV parametric maps, where the post-processing includes: (i) and (ii) the following: (i) partial volume correction, and (ii) Extraction fractions to estimate myocardial blood flow (MBF) at rest and during stress. d. Post-processing the resting and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR), wherein the artificial neural network is selected from the group consisting of a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network (CNN), a 1D convolutional neural network (1D-CNN), a recurrent neural network (RNN), a long-short-term memory recurrent neural network (LSTM-RNN), a gated recurrent unit (GRU) network, a generative adversarial network (GANs), deep machine learning, and / or combinations thereof.

[0066] Figure 1 is a flow diagram in which the model takes a blood input function and a time series at a voxel and outputs K1, k2, and TBV (total blood volume).

[0067] Figure 2 is a flow diagram in which the model takes the blood input function, the time series in the voxel, the distribution volume (DV), and outputs K1,TBV (total blood volume).

[0068] Figure 3 shows a flow diagram of model training. The model takes as input a blood input function, a voxel time series, and a distribution volume (DV), and outputs K1 and TBV (total blood volume). K1, TBV, and DV are input to a one-tissue compartment model to generate a predicted voxel time series. The mean squared error between the predicted voxel time series and the input voxel time series is calculated and fed back to the model to update the model parameters.

[0069] Figure 4 shows a comparison of the ground truth and AI-generated parametric maps. A shows the results of a patient's 82Rb stress scan. From left to right, the ground truth K1 parametric map generated by nonlinear least squares (NLS) is labeled K1 True, the K1 map predicted by the AI ​​model is labeled K1 Predicted (K1 Pred), the difference between K1 True and K1 Pred is labeled K1 Difference (K1 diff), and the correlation between the global images of K1 True and K1 Pred is shown. The solid line is the identity line, and the dotted line is the fit of the correlation. B shows the results of a patient's 82Rb resting scan. From left to right, the ground truth K1 parametric map generated by NLS (K1 True), the K1 map predicted by the AI ​​model (K1 Pred), the difference between K1 True and K1 Pred (K1 Difference), and the correlation between the global images of K1 True and K1 Pred are shown. The solid lines are identity lines and the dotted lines are correlation fits.

[0070] One embodiment of the present invention includes an image processing method for quantitative myocardial perfusion and myocardial flow reserve assessment, where consistent image quality is observed over a dose range of about 1 MBq to about 10,000 MBq of Rb-82.

[0071] One embodiment of the present invention includes an imaging method for assessing quantitative myocardial blood flow and myocardial blood flow reserve, where the imaging includes radiography, fluoroscopy, magnetic resonance imaging (MRI), computed tomography (CT), medical ultrasound or endoscopic ultrasound elastography, tactile imaging, thermography, medical photography, and nuclear medicine functional imaging techniques (e.g., positron emission tomography (PET), dynamic positron emission tomography, single photon emission computed tomography (SPECT), etc.), and / or combinations thereof.

[0072] Another embodiment of the present invention includes an image processing method for quantitative myocardial perfusion and myocardial flow reserve assessment, where the dose of administered contrast agent is calculated by an automated generation and injection system.

[0073] Yet another embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, wherein input signals are fed into a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network, and / or a long-short-term memory (LSTM) network to simultaneously predict the uptake rate (K1), the washout rate (k2), and the total blood volume (TBV), and an evaluation of the resulting images is performed to diagnose a disease state using a deep neural network, an artificial neural network, a deep machine learning, or a combination thereof.

[0074] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, comprising the following steps a to d for implementing and performing the test. a) producing a sufficient amount of Rb-82 by an automated elution system of a Sr-82 / Rb-82 radionuclide generator; b) administering the generated dose of Rb-82 to a patient; c) performing an appropriate imaging procedure to obtain a higher quality image of the small area; and d) performing an evaluation of the acquired images using deep neural networks, artificial neural networks, deep machine learning, convolutional neural networks, recurrent neural networks, long-short-term memory recurrent neural networks (LSTM-RNN), generative adversarial networks (GANs), gated recurrent unit (GRU) networks, and / or combinations thereof to diagnose a disease state.

[0075] One embodiment of the present invention includes an image processing method for assessing quantitative myocardial blood flow and myocardial blood flow reserve, the method further comprising administering a stress agent to the subject.

[0076] Another embodiment of the present invention includes an imaging method for assessing quantitative myocardial blood flow and myocardial blood flow reserve, wherein stress is induced by administering a stress agent selected from adenosine, adenosine triphosphate, regadenoson, dobutamine, dipyridamole, or exercise.

[0077] Furthermore, one embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, in which the subject weighs 1 to 300 kg, preferably 20 to 200 kg.

[0078] One embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, wherein the automated dose calculation further includes other parameters selected from the type of radioisotope, the half-life of the radioisotope, the generator life (activity remaining in the radioisotope generator), the generator yield, the injection time, the flow rate, the time elapsed from the generation of the radioisotope to the injection, the detector sensitivity of the scanning equipment, the scanner resolution, the type of camera or scanner, the imaging time, the camera sensitivity, the type of disease being diagnosed, the subject's condition or other special needs such as known allergies, cardiac function, liver function, renal function, supplementary diseases of the subject, medications, the type of imaging technique used such as PET, SPECT, CT, MRI, and / or combinations thereof.

[0079] Another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, and the automated generation and injection system includes a cabinet, a radioisotope generator, a dose calibrator, a computer, a controller, a display device, an activity detector, a cabinet, a cart, a waste bottle, a sensor, a shielding assembly, an alarm or alarm mechanism, tubing, a source vial, a diluent or eluent, a valve, or a combination thereof. The automated generation and injection system generates radionuclides from a generator / column disposed within the system. The radionuclide eluate is generated from the generator by eluting the generator with an appropriate eluent, such as saline, and is automatically administered by the system after radioactivity measurement. The dose is automatically calculated by the system based on input subject parameters. The system has the ability to calculate the flow rate and injection time according to the dose. The automated generation and injection system: 82 Sr / 82 It can consist of any radionuclide generator suitable for administration to a subject, such as a Rb generator.

[0080] One embodiment of the present invention includes an image processing method for quantitative myocardial perfusion and myocardial flow reserve assessment, where an automated generation and injection system is electronically or communicatively coupled to an image processing system that can issue a warning if image quality is subpar and a repeat administration or scan is required.

[0081] Another embodiment of the present invention includes an image processing method for quantitative myocardial perfusion and myocardial perfusion reserve assessment, wherein the automated generation and injection system is embodied in a portable (or mobile) cart that houses some or all of the following: generator, processor, pump, memory, patient line, bypass line, positron detector and / or calibrator, sensor, dose calibrator, activity detector, waste bottle, controller, display, and computer. The cart with the components for radioisotope generation and injection is mobile and can be moved to a patient location or center, hospital, etc. as needed.

[0082] Further, one embodiment of the present invention includes an image processing method for quantitative myocardial blood flow and myocardial blood flow reserve assessment, wherein the method of diagnosing / imaging blood perfusion or blood flow in a region of interest includes: inputting subject parameters into a radioisotope generation and injection system, automatically calculating the appropriate dose, generating a radionuclide from the automatic generation or injection system based on the required dose to be administered, administering the radionuclide to the subject in need, performing a PET or SPECT scan of the region of interest, automatically analyzing the image with computerized software, quantitatively assessing blood flow in the region of interest, generating an automated report of the assessment, and providing appropriate treatment options to the subject.

[0083] Another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, wherein the subject is a human subject. The human subject may be male or female. The subject's age may vary from 1 month to 120 years. The human subject may include neonatal, pediatric, adult, and / or geriatric populations.

[0084] Other embodiments of the present invention include image processing methods for quantitative myocardial blood flow and myocardial blood flow reserve assessment, and all numerical values ​​disclosed herein, when "about" is used, may vary by 1%, 2%, 5%, 10%, or up to 20%. This variation may apply to all numerical values ​​disclosed herein.

[0085] Another embodiment of the present invention includes an image processing method for quantitatively assessing myocardial blood flow and myocardial blood flow reserve, and the 3D parametric images of MBF generated by the present invention have better image quality and resolution. The present invention also provides recommended alerts to medical staff regarding the detection of coronary artery disease by analyzing the generated 3D parametric images of MBF and / or MFR.

[0086] Other embodiments of the present invention include image processing methods for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve, and calcium scoring is also recommended using 3D parametric MBF images generated by the present invention. The coronary artery calcium (CAC) score reflects the total area of ​​calcium deposits and calcium density. A score of 0 indicates no calcium is present in the heart, suggesting a low likelihood of future heart attacks. If calcium is present, a higher score indicates a higher risk of heart disease. To assess the accuracy and reproducibility of visual estimation of coronary artery calcium (CAC), positron emission tomography (PET), dynamic positron emission tomography (PET), hybrid positron emission tomography (PET), computed tomography (CT), and single-photon emission computed tomography (SPECT) / CT myocardial perfusion imaging (MPI) scans are performed.

[0087] Other embodiments of the present invention include an image processing method for assessing quantitative myocardial perfusion and / or myocardial perfusion reserve, wherein an image reconstruction algorithm is developed to improve image quality, and uses AI algorithms to improve the quality of image reconstruction, making image processing faster, and intended to reduce nuclear medicine dose during myocardial perfusion imaging (MPI) by up to 10 times.

[0088] Other embodiments of the present invention include image processing methods for AI models to generate blood flow parametric maps with high accuracy and in a time frame acceptable for clinical use, which may enable future clinical implementation. Each embodiment disclosed herein is applicable to each other disclosed embodiment. Accordingly, all combinations of the various elements described herein are within the scope of the present invention.

[0089] The present invention will be better understood by reference to the experimental data set forth below, but those skilled in the art will readily appreciate that the specific experiments detailed are merely illustrative of the invention as more fully described in the claims that follow. [Example]

[0090] Example 1 Patients were administered Rb-82. Scans of 20 subjects / patients (N=20) with a wide range of uptake defect severity on Rb-82 stress PET perfusion imaging were selected. Input signals were used to simultaneously predict the uptake rate (K1), k2, and total blood volume (TBV) using multilayer perceptrons (MLPs), artificial neural networks (ANNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long-short-term memory recurrent neural networks (LSTM-RNNs), gated recurrent unit (GRU) networks, generative adversarial networks (GANs), deep machine learning, and / or their combinations. Then, 3D parametric images of K1, k2, and TBV were combined to estimate MBF and / or MFR.

[0091] Example 2 Rb-82 was identified from a 2019 cardiac PET study administered to 40 patients (N=40) from two scanners (20 from GE Discovery 690 and 20 from GE Discovery 600) covering a wide range of defect severity for 82Rb stress PET. The Discovery 690 data were split into training / validation / test sets in a 60:20:20 ratio. All Discovery 600 data constituted a separate holdout test set. This study used a 196 x 196 x 98 mm pericardial area. 3 Image-derived arterial input functions (AIFs) and voxel time series / time-activity curves (TACs) in the region of the vein were used. Motion modeling was performed using a one-tissue compartment model (1TCM) with general nonlinear least squares (NLS) to generate reference parametric maps. The AIFs and voxel TACs were input into a convolutional / long-short-term memory neural network (CNN-LSTM) to predict K1 and TBV, as well as the associated predicted TAC. The AI ​​model was optimized to minimize the mean squared error between the input TAC and the predicted TAC (Figure 3). The results are shown below.

[0092] result The AI ​​model accurately predicted K1 and TBV, achieving an average R 2 The values ​​were 0.998 and 0.991 for the Discovery 690 and 0.995 and 0.997 for the holdout test set of the Discovery 600 (Figure 4). When generating a parametric map on a typical central processing unit (CPU), the conventional NLS method took an average of 89.1 minutes, while the AI-enabled model described in Example 2 took 7.21 seconds, 741 times faster.

[0093] These two embodiments of the AI ​​model of the present invention are capable of generating blood flow parametric maps with high accuracy and in a clinically usable time frame, making them suitable for future clinical implementation.

Claims

1. 1. An image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve, comprising: a. Pre-processing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion imaging (MPI) data; (ii) separating the value of voxel (i,j,k) at each time point ti (i=1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting a blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) K1, k ２ and the ratio of uptake rate to washout rate (K1 / k) to stabilize and improve the estimation of total blood volume (TBV) and subsequent myocardial blood flow measurement. ２ ) estimating the volume of distribution (DV) obtained by (vi) normalizing the data by dividing by the maximum value of the blood input function; and b. Evaluating the individual signals preprocessed in step (a) to generate parametric maps of K1 and TBV using an artificial neural network; and c. K1, k to estimate myocardial flow reserve (MFR) maps and / or coronary flow reserve (CFR) maps ２ and post-processing the parametric maps of TBV and myocardial blood flow at rest and stress. A method comprising:

2. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the image reconstruction of the array is a dynamic series of PET-reconstructed 3D tomographic image values ​​arranged over a number of time steps ti (i is 1 to N).

3. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the region of interest (ROI) can be set manually and / or automatically.

4. 2. An image processing method for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the data are normalized by dividing by the maximum value of the blood input function.

5. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the normalization of data by the value of the blood input function is between 0 and 1.

6. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the input signal is input to a multi-layer perceptron (MLP) and / or an artificial neural network (ANN) and / or a convolutional neural network and / or a long short-term memory (LSTM) network, and an uptake rate (K1), k ２ , and total blood volume (TBV) simultaneously.

7. 10. An image processing method for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve as claimed in claim 1, wherein regional blood flow and reserve values ​​are generated from images to highlight small regional blood flow defects.

8. 2. The image processing method for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the artificial neural network is selected from the group consisting of multi-layer perceptrons (MLPs), artificial neural networks (ANNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory recurrent neural networks (LSTM-RNNs), gated recurrent unit (GRU) networks, generative adversarial networks (GANs), deep machine learning, and / or combinations thereof.

9. 2. The image processing method for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the artificial neural network goes through multiple layers to estimate the volume of distribution (DV), the multiple layers being selected from the group consisting of initial layers, intermediate layers, final layers of the network, or a combination thereof.

10. 10. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 9, wherein the model is k ２ A method for predicting (washout rate) values.

11. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the method is performed voxel-wise using 1D signal CNN-LSTM to obtain K1, k ２ and a method for estimating total blood volume (TBV) to produce a more accurate myocardial blood flow (MBF) estimate.

12. 10. The imaging method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve of claim 1, wherein the imaging is characterized by administering Rb-82, O-15, N-13, Cu-62-PTSM, 99m-Tc-sestamibi, Tl-201, and / or combinations thereof.

13. 10. The image processing method for assessing quantitative myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the images are characterized by administering Rb-82 to highlight small regional blood flow defects during rest and stress PET perfusion imaging.

14. 2. The imaging method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the contrast agent or radionuclide is administered by an automated production and injection system and / or intravenous administration of radiopharmaceuticals produced by nuclear fission, neutron activation, cyclotron and / or generator.

15. 10. The imaging method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the automated radioisotope generation and injection system comprises a Rb-82 elution system.

16. 2. The image processing method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the obtained images are fitted to a one-tissue compartment model to determine myocardial blood flow reserve and / or coronary blood flow reserve by predicting a value of the ratio of myocardial blood flow stress to myocardial blood flow rest, and the evaluation of the obtained images is performed using a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory recurrent neural network (LSTM-RNN), a gated recurrent unit (GRU) network, deep machine learning and / or combinations thereof to diagnose a disease state.

17. 2. The imaging method for quantitatively assessing myocardial blood flow and / or myocardial blood flow reserve according to claim 1, wherein the imaging comprises positron emission tomography (PET), dynamic positron emission tomography, single photon emission computed tomography (SPECT), magnetic resonance imaging (MRI), computed tomography (CT), and / or combinations thereof.

18. 1. A myocardial image processing method for quantitatively assessing myocardial blood flow and myocardial reserve, comprising: a. Pre-processing images obtained with a rubidium-82 radiotracer, comprising: (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) separating the value of voxel (i,j,k) at each time point ti (i=1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting a blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) K1, k ２ and the ratio of uptake rate to washout rate (K / k) to stabilize and improve estimation of total blood volume (TBV) and subsequent myocardial blood flow measurements. ２ ) estimating the volume of distribution (DV) obtained by (vi) normalizing the data by dividing by the maximum value of the blood input function; and b. Apply the time series and blood input function at voxel (i, j, k) simultaneously to the artificial intelligence network to obtain the intake K1, k ２ and predicting TBV; c. K1, k ２ and post-processing the TBV parametric map, (i) partial volume correction; (ii) Extraction fractions to estimate resting and stress myocardial blood flow (MBF); and d. Post-processing resting and stress myocardial blood flow to estimate myocardial flow reserve (MFR) maps and / or coronary flow reserve (CFR) maps, including recommending coronary artery disease by analyzing the MBF and / or CFR maps; A method comprising:

19. 1. A myocardial image processing method for quantitatively assessing myocardial blood flow and myocardial reserve, comprising: a. Pre-processing the image, (i) Reconstructing dynamic cine 3D tomographic myocardial perfusion image (MPI) data; (ii) separating the value of voxel (i,j,k) at each time point ti (i=1 to N); (iii) optionally removing noise to improve the quality of the image; (iv) extracting a blood input function from the left ventricular blood cavity region of interest (ROI) or other arterial blood region of interest (ROI); (v) K1, k ２ and the ratio of uptake rate to washout rate (K / k) to stabilize and improve estimation of total blood volume (TBV) and subsequent myocardial blood flow measurements. ２ ) estimating the volume of distribution (DV) obtained by (vi) normalizing the data by dividing by the maximum value of the blood input function; and b. Apply the time series and blood input function at voxel (i, j, k) simultaneously to the artificial intelligence network to obtain the intake K1, k ２ and predicting TBV, with an average R ２ a step with a value between 0.9 and 1; c. Post-processing the K1 and TBV parametric maps, (i) partial volume correction, and (ii) Extraction fractions to estimate resting and stress myocardial blood flow (MBF); and d. Post-processing rest and stress myocardial blood flow to estimate myocardial flow reserve (MFR) and / or coronary flow reserve (CFR), the artificial neural network is selected from the group consisting of a multi-layer perceptron (MLP), an artificial neural network (ANN), a convolutional neural network (CNN) and / or a 1D convolutional neural network (1D-CNN), a recurrent neural network (RNN), a long short-term memory recurrent neural network (LSTM-RNN), a gated recurrent unit (GRU) network, a generative adversarial network (GANs), deep machine learning, and / or combinations thereof; A method comprising:

Images