US-12620494-B2 - Predicting adverse health risk based on CT scan images and AI-enabled cardiovascular structure volumetry

US12620494B2US 12620494 B2US12620494 B2US 12620494B2US-12620494-B2

Abstract

Embodiments of the invention relate generally to systems and methods for facilitating determining a risk of a patient for an adverse health outcome. In one embodiment, an inventive method includes using an AI-enabled volume calculator to estimate cardiac structure volume from CT scan images; and using a computer enabled risk calculator to, based at least in part on the estimated volume, determine a risk of a patient for an adverse health condition. In some embodiments, a system for detecting patients at risk of an adverse health condition includes a computer enabled volume calculator configured to facilitate assessing the volume of a cardiovascular structure based at least in part on a set of CT scan images; the system can include a computer enabled risk calculator configured to determine, based at least in part on the assessed volume, whether the patient is at risk for an adverse health condition.

Inventors

Morteza Naghavi

Assignees

Morteza Naghavi

Dates

Publication Date: 20260505
Application Date: 20230210

Claims (13)

1 . A computer-implemented method for determining a risk of cardiovascular disease from CT scan images using an artificial intelligence (AI), the method comprising: receiving one or more CT scan images containing a patient's heart via a user input device; preprocessing, by at least one processor, the one or more CT scan images to improve image quality for analysis; processing, by the least one processor, the preprocessed one or more CT scan images using a cardiovascular structure AI model to generate segmentation masks for each cardiovascular structure and distinguish blood inside the cardiovascular structure from the wall of the cardiovascular structure; refining the segmentation masks with a post-processing module to improve accuracy; generating, by the at least one processor, three-dimensional representation of the segmented cardiovascular structures and spatial visualization overlays of the segmented cardiovascular structures onto the original CT images using distinct color codes; identifying and delineating with voxel level resolutions, by the at least one processor, the boundaries of the three-dimensional segmented cardiovascular structures, including the left atrium (LA), the left ventricle (LV), the right atrium (RA), the right ventricle (RV), and the left ventricular mass (LVM), and quantifying volumes of each of the identified cardiovascular structures (LA, LV, RA, RV, LVM) based on three dimensional parameters of each of the identified cardiovascular structures; comparing, by the at least one processor, the quantified volumes to a set of reference values derived from a population database, wherein the reference values are stratified by at least one of age, gender, and body surface area; inputting the calculated volumes and patient-specific data by at least one of age, gender, and body surface area into a cardiovascular risk prediction model trained on clinical outcomes data and cardiovascular structure volumes to predict adverse health outcomes; computing, by the at least one processor and by using the cardiovascular risk prediction model, the risk of developing at least one adverse health condition selected from the group consisting of atrial fibrillation (AF), heart failure (HF), stroke, and cardiac mortality; and outputting, by an output medium, a report comprising a cardiovascular disease risk score adjusted by the at least one of age, gender, and body surface area.
2 . The method of claim 1 , wherein the one or more CT scan image is selected from the group consisting of contrast enhanced and non-contrast enhanced CT scan images, ECG-gated cardiac CT scan images, non-gated cardiac CT scan images, non-gated full chest CT scan images, low dose lung cancer screening CT scan images, and lung diagnostic CT scan images.
3 . The method of claim 1 further comprises a step of combining the volume of cardiovascular structures with one or more health related variables resulting in a multivariate composite index to determine the risk of future adverse cardiovascular conditions.
4 . The method of claim 1 is operating in a cloud infrastructure on a computing environment.
5 . The method of claim 1 further comprises steps for training the cardiovascular structure AI model, wherein the steps include: acquiring paired contrast-enhanced and non-contrast enhanced images of the heart chambers, aorta and pulmonary arteries including the pulmonary trunk; manually creating labels for each cardiac chamber using contrast-enhanced image as the guide; and co-registering or aligning the labeled contrast-enhanced images over the non-contrast enhanced images and creating a unified image as the input to train a UNET neural network for measuring the volume of cardiovascular structures.
6 . The method of claim 1 , wherein the cardiovascular disease includes atrial fibrillation, heart failure, stroke, coronary heart disease, cerebrovascular events, chronic obstructive pulmonary diseases (COPD), emphysema, ischemic heart disease, cardiovascular mortality, and all-cause mortality.
7 . A system for detecting a patient at risk of a cardiovascular disease, the system comprising: an input medium; at least one processor, a computer-readable medium storing computer-readable instructions; and an output medium, wherein the system receives, using the input medium, one or more non-invasive cardiac scan images of the patient's heart, wherein the one or more images are acquired from a modality selected from the group consisting of a computed tomography (CT) scan, an echocardiographic scan, and a magnetic resonance imaging (MRI) scan; preprocesses, using the least one processor, the one or more cardiac scan images to improve image quality for analysis; processes, using the at least one processor, the preprocessed images using a cardiovascular structure AI model to generate segmentation masks for each cardiovascular structure and distinguish blood inside the cardiovascular structure from the wall of the cardiovascular structure; refines the segmentation with a post-processing module to improve accuracy; generates, using the at least one processor, three-dimensional representation of the segmented cardiovascular structures and spatial visualization overlays of the segmented cardiovascular structure onto the original CT images, using distinct color codes; identifies and delineates with voxel level resolutions using the at least one processor the boundaries of the three-dimensional segmented cardiovascular structures, including the left atrium (LA), the left ventricle (LV), the right atrium (RA), the right ventricle (RV), and the left ventricular mass (LVM), and quantify volumes of each of the identified cardiovascular structures (LA, LV, RA, RV, LVM) based on three dimensional parameters of each of the identified cardiovascular structures; compares, using the at least one processor, the quantified volumes to a set of reference values derived from a population database, wherein the reference values are stratified by at least one of age, gender, and body surface area; computes, using the at least one processor, the adjusted volumes based on patient-specific data including at least one of the patient's age, gender, and body surface area; inputs the patient-specific adjusted volumes into a cardiovascular risk prediction model, the model having been trained on clinical outcome data correlating the volumes of cardiovascular structures with adverse health outcomes; computes, using the cardiovascular risk prediction model, the risk of developing at least one adverse health condition selected from the group consisting of atrial fibrillation (AF), heart failure (HF), stroke, and cardiac mortality; and outputs, using the output medium, a report comprising the risk of developing at least one adverse health condition selected from the group consisting of atrial fibrillation (AF), heart failure (HF), stroke, and cardiac mortality.
8 . A system for detecting a patient at risk of developing atrial fibrillation (AF), the system comprising: an input medium; at least one processor; a computer-readable medium storing computer-readable instructions; and an output medium, wherein the system receives, using the input medium, one or more non-invasive cardiac scan images of the patient's heart, wherein the one or more images are acquired from a modality selected from the group consisting of a computed tomography (CT) scan, an echocardiographic scan, and a magnetic resonance imaging (MRI) scan; preprocesses, using the at least one processor, the one or more noninvasive scan images to improve image quality for analysis; processes, using the at least processor, the preprocessed images using a cardiovascular structure AI model to generate a segmentation mask of the left atrium (LA) and distinguish blood inside LA from the wall of LA; refines the segmentation with a post-processing module to improve accuracy; generates, using the at least one processor, a three-dimensional representation of the segmented LA and spatial visualization overlay of the segmented LA on the original images, using a distinct color code; identifies and delineates with voxel level resolutions the boundaries of the segmented LA and quantify the volume based on three-dimensional parameters of the segmented LA; compares, using the at least one processor, the quantified LA volume to a set of reference values derived from the population database, wherein the reference values are stratified by at least one of age, gender, and body surface area; compute, using the at least one processor, the adjusted LA volume based on patient-specific data including at least one of the patient's age, gender, and body surface area; inputs the patient-specific adjusted LA volume into an AF risk prediction model, the model having been trained on clinical outcome data correlating LA volume with incidence of atrial fibrillation in a population; computes an AF risk score for the patient using the AF risk prediction model, wherein the AF risk score is determined based on the patient's adjusted LA volume and wherein a larger LA volume (a higher percentile) corresponds to a higher risk of AF; and outputs a report comprising the patient's atrial fibrillation risk score via the output medium.
9 . The system of claim 8 further comprising a computer enabled health adviser configured, based on the LA volume and a patient's age, gender, ethnicity, body surface size, and other health related conditions, for recommending a cardiac monitoring device for monitoring episodes of AF and/or for alerting patients to take preventive actions against a future cerebrovascular event.
10 . The system of claim 9 , wherein the cardiac monitoring device is one from the group comprising: implantable ECG devices, wearable ECG devices, ECG patches, ECG embedded blood pressure cuffs, photoplethysmography (PPG) devices, and wearables devices capable of detecting AF and alerting the patient.
11 . The system of claim 8 , wherein the risk calculator is further configured to determine, based at least in part on one or more health related variables resulting in a multivariate composite index to estimate the risk of developing AF and stroke.
12 . A system for detecting a patient at risk of developing heart failure with reduced ejection fraction (HFrEF) versus heart failure with preserved ejection fraction (HFpEF), comprising an input medium; at least one processor; a computer-readable medium storing computer-readable instructions; and an output medium, wherein the system receives, by the input medium, a plurality of non-invasive cardiac scan images of the patient's heart, wherein the images are obtained from at least one modality selected from the group consisting of a computed tomography (CT) scan, an echocardiographic scan, and a magnetic resonance imaging (MRI) scan; preprocesses, by the at least one processor, the one or more cardiac scan images to enhance image quality for analysis; processes, by the at least one processor, the preprocessed images using a cardiovascular structure AI model to generate segmentation masks for each cardiovascular structure and distinguish blood inside the cardiovascular structure from the wall of the cardiovascular structure; refines the segmentation with a post-processing module to improve accuracy; generates three-dimensional representation of the segmented cardiovascular structures and spatial visualization overlays of the segmented cardiovascular structure onto the original images, using distinct color codes; identifies and delineates with voxel level resolutions the boundaries of the three-dimensional segmented cardiovascular structures including the left atrium (LA), the left ventricle (LV), the right atrium (RA), the right ventricle (RV), and the left ventricular wall (LVW), and quantify volumes of the segmented cardiac structures (including at least the LA, LV, RA, RV, and LVW) based on the segmentation data; adjusts, by the at least one processor, each calculated chamber volume based on patient-specific parameters including at least the patient's age, gender, and body surface area, thereby normalizing the volumes for the patient's body size and demographics; compares the quantified volumes to a set of reference values derived from the population database, wherein the reference values are stratified by at least one of age, gender, and body surface area; adjusts the quantified volumes based on patient-specific data including at least one of the patient's age, gender, and body surface area (BSA); inputs the patient-specific adjusted volumes data into a heart failure risk prediction model, the model being trained on clinical data to correlate patterns of cardiac chamber volumes with the likelihood of HFrEF or HFpEF; computes a heart failure risk score for the patient using the heart failure risk prediction model, wherein the risk outcome indicates a likelihood of the patient developing HFrEF versus HFpEF, and wherein an abnormally elevated LV volume is indicative of a higher risk of HFrEF while a relatively normal an abnormally elevated LV mass and relatively normal LV volume (with other volume indicators, such as an enlarged LA or thickened LV wall, if present) is indicative of a higher risk of HFpEF; and outputs, by the output medium, a report identifying the patient's heart failure risk, including an indication of whether the patient is at greater risk for HFrEF or for HFpEF based on the volumetric analysis.
13 . The system of claim 12 further comprises a computer enabled health adviser configured to, based at least on patient's age, gender, ethnicity, body surface area, cardiometabolic risk factors, past medical history, and other health related conditions, facilitate recommending a treatment plan to take preventive actions against a future cardiovascular event.

Description

RELATED APPLICATIONS This application claims priority to U.S. provisional patent application 63/414,546, filed on Oct. 9, 2022, which is hereby incorporated in its entirety herein by reference. BACKGROUND Field Embodiments of the invention relate generally to system and methods for determining the risk of a person for an adverse health condition. In particular, embodiments of the invention are directed to systems and methods for using non-invasive images to train an artificial intelligence model to facilitate estimating the volume of a cardiovascular structure. Based at least in part on the estimated volume, a risk of an adverse health condition can be determined. DESCRIPTION OF RELATED ART For coronary artery disease certain screening and diagnostic tools are known, such as coronary artery calcium (CAC) score and coronary CT angiography. And for predicting atrial fibrillation (AF) and heart failure (HF) it is known to use, for example, CHARGE-AF and brain natriutic peptide (BNP). CHARGE AF is an epidemiological risk calculator. BNP is more precise than CHARGE-AF but it is not specific to left atrial and ventricular function. AF is the most common sustained arrhythmia and is associated with an increased risk of stroke and cardiovascular mortality. In the United States, at least 3 to 6 million people have AF. It is predicted that the number of AF patients will increase to over 12 million cases by 2030, imposing a significant economic burden with projected healthcare expenses of $260 million. In Europe, prevalent AF in 2010 is about 9 million among individuals older than 55 years and is expected to reach 14 million by 2040. It is estimated that by 2050 at least in 72 million individuals in Asia will be diagnosed with AF, and about 3 million with AF-related strokes. This presents a public health crisis especially for the growing elderly population in coming decades. Medicare services costs are significantly higher among AF patients than non-AF patients; therefore early treatment is critical to limit the disease burden imposed by AF. The adverse social and public health effects of HF are even worse. It is estimated that by 2030, more than 8 million Americans will have HF. And the total direct medical costs of HF are expected to rise from $21 billion to $53 billion. The 5-year survival rates of AF are of concern. Without proper treatment, 51% of AF patients will die within five years. Although the economic burden posed by AF and HF are critical in light of the increasing healthcare costs, early detection tools and preventive interventions for pre-AF and pre-HF patients are currently unavailable. One report shows that 96,860 strokes occurred within 1 year among patients with AF, with an associated total direct lifetime cost of nearly $8 billion. Of these costs, $2.6 billion in direct costs are incurred during the first year after the stroke. HF poses an even greater threat to US healthcare system. Given the rising rates of hospitalization and rehospitalization, HF is associated with a significant cost burden. Approximately 1% to 2% of the total US health care budget is spent on HF, and half of that is attributable to late diagnosis leading to inpatient admissions for HF. This challenge presents a great opportunity to make an impact on the healthcare system by early detection and interventions of subclinical HF and AF. Currently, BNP and CHARGE-AF are the only available tools for early detection of high-risk patients for AF and HF. A combination of high-risk CHARGE-AF and a 7-day ECG patch has been reported. CHARGE-AF is an epidemiological risk calculator that can be useful as a population based measure, but it is not preferred as applied to individual patients as needed in a physician's office. A more direct assessment is preferred such as by imaging the cardiac chambers where AF happens. It is known to calculate a CHARGE-AF score: 0.508×age (5 year increments)+0.248×height (10 cm increments)+0.115×weight (15 Kg increments)+0.197×systolic blood pressure (20 mm Hg increments)−0.101×diastolic blood pressure (10 mm Hg increments)+0.359×current smoker+0.349×antihypertensive medication+0.237×diabetes+0.701×congestive heart failure+0.496×myocardial infarction. See, “Simple risk model predicts incidence of atrial fibrillation in a racially and geographically diverse population: the CHARGE-AF consortium.” J Am Heart Assoc. 2013 Mar. 18; 2 (2): e000102. doi: 10.1161/JAHA.112.000102. PMID: 23537808; PMCID: PMC3647274. It is known to use manual measurements of left ventricle chamber in a single-slice to predict HF. However, rapid and accurate acquisition of whole heart volume parameters is challenging. Even though semi-automated delineation and quantification of cardiovascular structures can be useful in CT images, currently known methods still require a significant degree of manual modifications, which is time-consuming and may increase inter-/intra-observer variability. Over a billion will die from cardiovascular disease,