Artificial Intelligence-Enhanced Screening for Cardiac Amyloidosis by Electrocardiogram Recording

Abstract

Provided herein are methods, systems, and computer program products for detection of cardiac amyloidosis (CA), the methods, systems, and computer program products including receiving voltage-time data for multiple leads of an electrocardiograph of a subject, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving an indication of the presence or absence of CA in the subject from the pre-trained learning system.

Description

[Technical field]

[0001] Related Applications This application claims the benefit of priority to U.S. Provisional Application No. 63 / 336,493, filed April 29, 2022, which is incorporated by reference herein in its entirety.

[0002] Embodiments of the present disclosure relate to methods for the detection and treatment of cardiac amyloidosis. [Background technology]

[0003] Cardiac amyloidosis (CA), once thought to be rare and universally fatal, is now recognized as an important cause of heart failure, especially in patients with preserved ejection fraction. Although therapeutic advances have led to significant improvements in outcomes, survival is hampered by life-threatening diagnostic delays. While over 30 proteins can misfold to cause amyloidosis, the two major types involve the heart: light chain-associated amyloid (AL), which results from a clonal plasma cell disorder in the bone marrow, and transthyretin amyloid (ATTR), which is associated with misfolding of transthyretin produced by the liver. ATTR can result from either inherited mutations in the transthyretin gene (ATTRv) or "wild-type" (genetically normal) transthyretin deposits (ATTRwt). Treatments are available for both AL and ATTR amyloidosis and are improving rapidly. 4,5 If untreated, cardiac AL amyloidosis is rapidly progressive and fatal. Despite improvements in chemotherapy, approximately 25% of patients with AL amyloidosis die from advanced cardiac disease within 6 months of diagnosis. Cardiac ATTR amyloidosis progresses more slowly, whereas median survival is approximately 3.5 years in ATTRwt and somewhat shorter in ATTRv. Summary of the Invention [Problem to be solved by the invention]

[0004] Thus, embodiments of the invention disclosed herein include algorithms applied to electrocardiograms (ECGs), a non-invasive procedure, in the diagnostic workup of CA for detection and stratification of patients based on risk of CA, thereby enabling earlier diagnosis and intervention. [Means for solving the problem]

[0005] According to embodiments of the present disclosure, a method and computer program product for the detection of cardiac amyloidosis (CA) are provided.

[0006] In some aspects of the invention, a method is disclosed herein that includes receiving voltage-time data of a subject, where the voltage-time data includes voltage data for multiple leads of an electrocardiograph, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving from the pre-trained learning system an indication of the presence or absence of cardiac amyloidosis in the subject.

[0007] Aspects of the invention disclosed herein also include a system comprising an electrocardiograph having multiple leads; and a computing node comprising a computer readable storage medium embodied with program instructions, the program instructions being executable by a processor of the computing node such that the processor performs a method comprising receiving voltage-time data of a subject from the echocardiograph, the voltage-time data including voltage data for the multiple leads; generating a feature vector from the voltage-time data; providing the feature vector to a pre-trained learning system; and receiving an indication of the presence or absence of cardiac amyloidosis in the subject from the pre-trained learning system.

[0008] In a particular aspect of the invention, disclosed herein is a computer program product for detection of cardiac amyloidosis, comprising a computer readable storage medium embodied with program instructions, the program instructions being executable by a processor such that the processor performs a method including receiving voltage-time data of a subject from an echocardiograph, the voltage-time data including voltage data for the plurality of leads, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving from the pre-trained learning system an indication of the presence or absence of cardiac amyloidosis in the subject. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram of a system for detecting CA according to an embodiment of the present disclosure. [Diagram 2] 1 is a flow chart illustrating a method for detecting cardiac amyloidosis according to an embodiment of the present disclosure. [Diagram 3] 1 illustrates a computing node according to an embodiment of the present disclosure. [Figure 4] Schematic summary of the study design (left), a clinical example of early detection of cardiac amyloid using AI-enhanced ECG prior to clinical diagnosis (center, top), a normal 12-lead ECG at the time the model predicted cardiac amyloid (center, bottom), and model performance (right). [Diagram 5] Figure 12-lead model receiver operating characteristic curves. Test set receiver operating characteristic curves for a 12-lead network trained to discriminate between controls and controls with cardiac amyloidosis. Using an optimal probability threshold of 0.485, the sensitivity is 0.84 and the specificity is 0.85. AUC, area under the receiver operating characteristic curve. [Figure 6] A calibration curve is shown. Calibration compares the probability of amyloid predicted by the model (x-axis) to the observed amyloid rate in the population (y-axis). The dashed line represents a perfectly calibrated model. [Figure 7]A Violin plot of predicted amyloid probability in the control group is shown. The distribution of model predictions is for controls only. The white dot is the median probability and the thick bar is the interquartile range. The ends of the thin needle are the neighboring values ​​above and below. Observations above and below these values ​​are often suspected outliers. The width of the plot represents the number of observations. [Figure 8] Violin plots of predicted amyloid probability in cardiac amyloidosis patients with at least one electrocardiogram (ECG) >6 months prior to diagnosis are shown. See Fig. 4 for explanation of violin plot components. Here, available ECGs were categorized into 6-month groups up to 5 years prior to clinical diagnosis. For patients with multiple ECGs in each bin, the median probability was used. "n" represents the number of patients with ECGs in the corresponding time window. Light chain (AL) myocardial amyloid is shown in Fig. 8A and wild-type transthyretin amyloid (ATTRwt) in Fig. 8B. [Figure 9] In Figures 9A and 9B, artificial intelligence (AI)-enhanced electrocardiogram (ECG) for early detection of cardiac amyloidosis (CA): model development, performance, and clinical example. In a 67-year-old male with unexplained heart failure (HF), the AI ​​model predicted amyloidosis when the echocardiogram was not interpreted as consistent with amyloid and showed only a borderline increase in left ventricular wall thickness. The 12-lead ECG was interpreted as normal when the AI ​​model first predicted cardiac amyloidosis. AL, light chain amyloid. [Figure 10] 1 shows a model architecture used with input consisting of raw ECG waveform data and a binary output that predicts the presence or absence of CA. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Patients with systemic amyloidosis are commonly evaluated by multiple providers before a diagnosis is established, often requiring transfer to a national center. Cardiologists commonly evaluate these patients, but in one study, the correct diagnosis is established in only 18.7% of cases. Early symptoms are vague and may be attributable to other causes, necessitating a high index of suspicion. Advances in cardiac imaging have improved diagnosis, but classic findings are not always present or may go unrecognized, especially if amyloidosis has not been suggested by the referring provider. Because cardiac involvement in systemic amyloidosis is the most important determinant of survival, there is a critical need for early diagnosis to facilitate timely and effective treatment.

[0011] Classical ECG findings of AL CA include low voltage and pseudoinfarction patterns present in approximately 45% of patients, but neither normal voltage nor ECG criteria for left ventricular hypertrophy (LVH) exclude the diagnosis. Despite more severe myocardial infiltration, low voltage is present in only approximately 25% of ATTRwt patients. Conduction system disease and ST-segment and T-wave abnormalities are common in both types of CA but are nonspecific. Artificial intelligence (AI)-driven models have been reported to enable identification of occult and impending disease from ECG findings that appear nonspecific or insignificant to expert reviewers. 12

[0012] Disclosed herein is the development of an artificial intelligence (AI)-based tool for detecting cardiac amyloidosis (CA) from standard 12-lead electrocardiograms (ECGs). A novel AI electrocardiogram (ECG) network was developed to test an AI-driven model for early detection of CA. Convolutional neural networks provide a comprehensive approach to analyze and interpret the vast amount of data generated by a single ECG. The algorithm was developed using 12-lead ECG data collected from 2541 patients with light chain or transthyretin CA seen at Mayo Clinic from 2000 to 2019. Cases were age- and sex-matched nearest neighbors and 2454 controls. A subset of 2997 (60%) cases and controls was used to train a deep neural network to predict the presence of CA in an internal validation set (n=999; 20%) and a randomly selected holdout test set (n=999; 20%) (Figure 4). Experiments were performed using single-lead and 6-lead ECG subsets. Furthermore, because smartphone-compatible electrodes enable point-of-care diagnostics with single-lead and 6-lead options, a compatible network was developed and tested.

[0013] All models used voltage-time information from the 12-lead ECG as input. Modeling techniques investigated included convolutional neural networks with different structures, such as using all 12 leads as a single input, each lead converted to a spectrogram, a group of 6 leads as a separate input, a single lead as an input, and combinations of these methods. Thus, embodiments of the invention disclosed herein include the use of at least 1 to 12 leads. Furthermore, disease may also include an underlying genetic component, supported by detection of these diagnostic signals prior to diagnosis. The methods disclosed herein may be combined with gene panels to provide additional specificity and sensitivity. Similarly, such methods may be used with gene and biomarker panels (e.g., cardiac biomarkers B-type natriuretic peptide (BNP), N-terminal (NT) prohormone BNP (NT-proBNP), troponin) for early and accurate detection of disease and to detect novel biomarkers of disease.

[0014] Thus, in some aspects of the invention, disclosed herein is a method that includes receiving voltage-time data of a subject, the voltage-time data including voltage data for multiple leads of an electrocardiograph, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving an indication of the presence or absence of cardiac amyloidosis in the subject from the pre-trained learning system. Generating the feature vector may include generating a spectrogram based on the voltage data for the multiple leads. In some embodiments, generating the feature vector includes grouping the voltage data for the multiple leads into multiple subsets.

[0015] In some embodiments, such methods further include receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector. In some such embodiments, the method further includes receiving genomic information of the subject. Generating the feature vector may include adding the genomic information to the feature vector. Without being bound to any particular methodology or theory, the genomic data may be derived from a biological sample from an inherited ATTR patient carrying a mutation. Such mutations are known in the art and may include, without limitation or exclusion, mutations V30M, Y114C, G47R, S50I, T49S, F33V, A45T, E89K, E89Q, and V122I. In some such embodiments, the learning system comprises a convolutional neural network. Such a convolutional neural network may include at least one residual connection.

[0016] In some embodiments, the subject's voltage-time data is received from an electrocardiograph. In further embodiments, the subject's voltage-time data is received from an electronic medical record.

[0017] In some embodiments, the method further includes providing the indication to an electronic health record system for storage in a health record associated with the subject. In some embodiments, the method further includes providing the indication to a computing node for display to a user.

[0018] In some embodiments of the methods disclosed herein, the feature vector comprises a matrix having a number of rows and a number of columns, the number of rows corresponding to a time dimension and the number of columns corresponding to a spatial dimension. In some such embodiments, each of the number of rows corresponds to one of the number of leads and each of the number of columns corresponds to a timestamp. In some embodiments, the time dimension has a resolution of 500 Hz. The convolutional neural network disclosed herein may comprise at least nine convolutional blocks and two fully connected blocks.

[0019] Referring now to FIG. 1, a system for detecting cardiac amyloidosis according to an embodiment of the present disclosure is shown. As outlined above, in various embodiments, patient information including electrocardiogram (ECG) data is provided to a learning system to determine the presence of cardiac amyloidosis. Accordingly, aspects of the invention disclosed herein also include a system comprising an electrocardiograph with multiple leads, and a computing node comprising a computer-readable storage medium embodied with program instructions, the program instructions being executable by a processor of the computing node, such that the processor executes a method including receiving voltage-time data of a subject from the echocardiograph, the voltage-time data including voltage data of the multiple leads, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving an indication of the presence or absence of cardiac amyloidosis in the subject from the pre-trained learning system. Generating the feature vector may include generating a spectrogram based on the voltage data of the multiple leads. In some embodiments, generating the feature vector includes grouping the voltage data of the multiple leads into a plurality of subsets.

[0020] In some embodiments, such a system further includes receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector. In some such embodiments, the system further includes receiving genomic information of the subject. Generating the feature vector may include adding the genomic information to the feature vector. Without being bound to any particular methodology or theory, the genomic data may be derived from a biological sample from an inherited ATTR patient carrying a mutation. Such mutations are known in the art and may include, without limitation or exclusion, mutations V30M, Y114C, G47R, S50I, T49S, F33V, A45T, E89K, E89Q, and V122I. In some such embodiments, the learning system comprises a convolutional neural network. Such a convolutional neural network may include at least one residual connection.

[0021] In some embodiments, the subject's voltage-time data is received from an electrocardiograph. In further embodiments, the subject's voltage-time data is received from an electronic medical record.

[0022] In some embodiments, the system further includes providing the indication to an electronic health record system for storage in a health record associated with the subject. In some embodiments, the system further includes providing the indication to a computing node for display to a user.

[0023] In some embodiments of the systems disclosed herein, the feature vector comprises a matrix having a number of rows and a number of columns, the number of rows corresponding to a time dimension and the number of columns corresponding to a spatial dimension. In some such embodiments, each of the number of columns corresponds to one of the number of leads, and each of the number of columns corresponds to a timestamp. In some embodiments, the time dimension has a resolution of 500 Hz. In some embodiments, the convolutional neural network comprises at least 9 convolutional blocks and 2 fully connected blocks.

[0024] Patient data may be received from an electronic health record (EHR) 101. An electronic health record (EHR) or electronic medical record (EMR) may refer to an organized collection of patient and population health information stored electronically in digital format. These records may be shared between different medical settings. Records may be shared through networked enterprise-wide information systems or other information networks and exchanges. An EHR may contain a variety of data, including demographics, medical history, medications and allergies, immune status, laboratory results, radiology images, vital signs, personal statistics such as age and weight, and billing information. An EHR system may be designed to store data and capture a patient's condition over time. In this way, there is no need to track a patient's previous paper medical records.

[0025] Electrocardiogram (ECG) data may be received directly from an electrocardiogram recording device 102. In an exemplary 12-lead ECG, ten electrodes are placed on the patient's limbs and the surface of the chest. The overall magnitude of the heart's electrical potential is then measured from twelve different angles (leads) and recorded over a period of time (usually 10 seconds). In this way, the overall magnitude and direction of the heart's electrical depolarization is captured at each instant throughout the cardiac cycle.

[0026] Additional data stores 103 may contain additional patient information as described herein. Suitable data stores include databases, flat files, and other structures known in the art.

[0027] It may be appreciated that the ECG data may be stored in the EHR for later retrieval. It may also be appreciated that the ECG data may be cached rather than delivered directly to the learning system for further processing.

[0028] The learning system 104 receives patient information from one or more of the EHR 101, the ECG 102, and the additional data store 103. As noted above, in some embodiments the learning system comprises a convolutional neural network. In various embodiments, inputs to the convolutional neural network include voltage-time information, the ECG, which in some embodiments is paired with additional patient information, such as demographic or genetic information.

[0029] The learning system 104 may be pre-trained using appropriate population data as described in the Examples to generate an indication of the presence or absence of cardiac amyloidosis. In some embodiments, the indication is binary. In some embodiments, the indication is a probability value indicating the likelihood of cardiac amyloidosis given the input patient data.

[0030] In some embodiments, the learning system 104 provides an indication of cardiac amyloidosis for storage as part of the EHR. In this manner, a computer-aided diagnosis is provided that can be referenced by a clinician. In some embodiments, the learning system 104 provides the indication of cardiac amyloidosis to a remote client 105. For example, the remote client may be a health app, a cloud service, or another consumer of diagnostic data. In some embodiments, the learning system 104 is integrated into an ECG machine to provide immediate feedback to the user during testing.

[0031] In some embodiments, a feature vector is provided to the learning system. Based on the input features, the learning system generates one or more outputs. In some embodiments, the output of the learning system is a feature vector.

[0032] In some embodiments, the learning system comprises an SVM. In other embodiments, the learning system comprises an artificial neural network. In some embodiments, the learning system is pre-trained using training data. In some embodiments, the training data is retrospective data. In some embodiments, the retrospective data is stored in a data store. In some embodiments, the learning system may be further trained by manual curation of previously generated outputs.

[0033] In some embodiments, the learning system is a trained classifier. In some embodiments, the trained classifier is a random decision forest. However, it can be appreciated that a variety of other classifiers are suitable for use with the present disclosure, including linear classifiers, support vector machines (SVMs), or neural networks such as recurrent neural networks (RNNs).

[0034] Suitable artificial neural networks include, but are not limited to, feedforward neural networks, radial basis function networks, self-organizing maps, learning vector quantization, recurrent neural networks, Hopfield networks, Boltzmann machines, echo state networks, long short term memories, bidirectional recurrent neural networks, hierarchical recurrent neural networks, probabilistic neural networks, modular neural networks, associative neural networks, deep neural networks, deep belief networks, convolutional neural networks, convolutional deep belief networks, large memory search neural networks, deep Boltzmann machines, deep stacking networks, tensor deep stacking networks, spike and slab restricted Boltzmann machines, composite hierarchical deep models, deep coding networks, multi-layer kernel machines, or deep Q networks.

[0035] In machine learning, a convolutional neural network (CNN) is a type of feed-forward artificial neural network applicable to the analysis of visual images and other natural signals. A CNN consists of an input layer, an output layer, and multiple hidden layers. The hidden layers of a CNN typically consist of a convolutional layer, a pooling layer, a fully connected layer, and a normalization layer. A convolutional layer applies a convolution operation to the input and passes the result to the next layer. The convolutions emulate the response of an individual neuron to a stimulus. Each convolutional neuron processes data only for its receptive field.

[0036] The convolution operation allows for a reduction in free parameters compared to fully connected feedforward networks. In particular, tiling a given kernel allows for learning a fixed number of parameters independent of the image size. This in turn reduces the memory footprint of a given network.

[0037] The parameters of a convolutional layer consist of a set of learnable filters (or kernels) that have small receptive fields but extend across the full depth of the input volume. During a forward pass, each filter is convolved across the width and height of the input volume, computing dot products between the filter's entries and the input, generating a 2D activation map for that filter. As a result, the network learns a filter to fire when it detects some particular type of feature at some spatial location in the input.

[0038] In an example convolution, the kernel is a set of weights w 1 ...w 9 It may be understood that the sizes provided here are merely exemplary and that any kernel dimensions may be used as described herein. A kernel is applied to each tile of an input (e.g., an image). The result of each tile is an element of a feature map. It may be understood that multiple kernels may be applied to the same input to generate multiple feature maps.

[0039] The full output volume of a convolutional layer is formed by stacking the feature maps of all kernels, so every entry in the output volume can also be interpreted as the output of a neuron that looks at a small region in the input and shares parameters with neurons in the same feature map.

[0040] Convolutional neural networks may be implemented on a variety of hardware, including hardware CNN accelerators and GPUs.

[0041] 2, a flow chart illustrating a method for detecting cardiac amyloidosis according to an embodiment of the present disclosure is provided. At 201, voltage-time data is received for a subject. The voltage-time data includes voltage data for multiple leads of an electrocardiograph. At 202, a feature vector is generated from the voltage-time data. At 203, the feature vector is provided to a pre-trained learning system. At 204, an indication of the presence or absence of cardiac amyloidosis in the subject is received from the pre-trained learning system.

[0042] 3, a schematic diagram of an example computing node is shown. Computing node 10 is merely one example of a suitable computing node and is not intended to suggest any limitation as to the scope of use or functionality of the embodiments described herein. Nevertheless, computing node 10 may implement and / or perform any of the functionality described above.

[0043] Computing node 10 includes computer system / server 12 that operates in numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations that may be suitable for use with computer system / server 12 include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable appliances, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems or devices.

[0044] The computer system / server 12 may be described in the general context of computer system executable instructions, such as program modules, executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The computer system / server 12 may also be practiced in a distributed cloud computing environment where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media, including memory storage devices.

[0045] 3, computer system / server 12 within computing node 10 is shown in the form of a general-purpose computing device. Components of computer system / server 12 may include, but are not limited to, one or more processors or processing units 16, a system memory 28, and a bus 18 that couples various system components including system memory 28 to processor 16.

[0046] Bus 18 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example and not limitation, such architectures include Industry Standard Architecture (ISA) bus, MicroChannel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, Peripheral Component Interconnect (PCI) bus, Peripheral Component Interconnect Express (PCIe), and Advanced Microcontroller Bus Architecture (AMBA).

[0047] Computer system / server 12 typically includes a variety of computer system readable media. Such media may be any available media that can be accessed by computer system / server 12 and includes both volatile and nonvolatile media, removable and non-removable media.

[0048] The system memory 28 may include computer system readable media in the form of volatile memory such as random access memory (RAM) 30 and / or cache memory 32. The computer system / server 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be provided for reading from and writing to a non-removable, non-volatile magnetic medium (not shown, typically referred to as a "hard drive"). Although not shown, a magnetic disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk") and an optical disk drive for reading from and writing to a removable, non-volatile optical disk, such as a CD-ROM, DVD-ROM, or other optical medium, may be provided. In such a case, each may be connected to the bus 18 by one or more data media interfaces. As further shown and described below, the memory 28 may include at least one program product having a set of program modules (e.g., at least one of the program modules) configured to perform the functions of an embodiment of the present disclosure.

[0049] A program / utility 40 having a set of program modules 42 (at least one of the program modules 42) may be stored in memory 28, as well as an operating system, one or more application programs, other program modules, and program data, by way of example and not limitation. Each of the operating system, one or more application programs, other program modules, and program data, or any combination thereof, may include an implementation of a networking environment. The program modules 42 generally perform the functions and / or methodologies of the embodiments described herein.

[0050] The computer system / server 12 may also communicate with one or more external devices 14, such as a keyboard, pointing device, display 24, one or more devices that allow a user to interact with the computer system / server 12, and / or any device (e.g., network card, modem, etc.) that allows the computer system / server 12 to communicate with one or more other computing devices. Such communication may occur via an input / output (I / O) interface 22. Additionally, the computer system / server 12 may communicate with one or more networks, such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet), via a network adapter 20. As shown, the network adapter 20 communicates with other components of the computer system / server 12 via a bus 18. Although not shown, it should be understood that other hardware and / or software components may be used with the computer system / server 12. Examples include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data archive storage systems, and the like.

[0051] The present disclosure may be embodied as a system, a method, and / or a computer program product. For example, in some aspects or the present invention, a computer program product for detecting cardiac amyloidosis is provided herein, the computer program product comprising a computer readable storage medium embodied with program instructions, the program instructions being executable by a processor, whereby the processor executes a method including receiving voltage-time data of a subject from an echocardiograph, the voltage-time data including voltage data of a plurality of leads, generating a feature vector from the voltage-time data, providing the feature vector to a pre-trained learning system, and receiving an indication of the presence or absence of cardiac amyloidosis in the subject from the pre-trained learning system. Generating the feature vector may include generating a spectrogram based on the voltage data of the plurality of leads. In some embodiments, generating the feature vector includes grouping the voltage data of the plurality of leads into a plurality of subsets.

[0052] In some embodiments, such computer program products further include receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector. In some such embodiments, the computer program further includes receiving genomic information of the subject. Generating the feature vector may include adding the genomic information to the feature vector. Without being bound to any particular methodology or theory, the genomic data may be derived from a biological sample from an inherited ATTR patient carrying a mutation. Such mutations are known in the art and may include, without limitation or exclusion, mutations V30M, Y114C, G47R, S50I, T49S, F33V, A45T, E89K, E89Q, and V122I. In some such embodiments, the computer program product comprises a convolutional neural network. Such a convolutional neural network may include at least one residual connection.

[0053] In some embodiments, the subject's voltage-time data is received from an electrocardiograph. In further embodiments, the subject's voltage-time data is received from an electronic medical record.

[0054] In some embodiments, the computer program product further includes providing a display to an electronic health record system for storage in a health record associated with the subject. In some embodiments, the computer program product further includes providing a display to a computing node for display to a user.

[0055] In some embodiments of the computer program product disclosed herein, the feature vector may include a matrix having a number of rows and a number of columns, where the number of rows corresponds to a time dimension and the number of columns corresponds to a space dimension. In some such embodiments, each of the number of columns corresponds to one of the number of leads, and each of the number of columns corresponds to a timestamp. In some embodiments, the time dimension has a resolution of 500 Hz. In some embodiments, the convolutional neural network comprises at least 9 convolutional blocks and 2 fully connected blocks.

[0056] Computer program products provided herein may include a computer readable storage medium (or multiple computer readable storage media) having computer readable program instructions for causing a processor to perform aspects of the present disclosure.

[0057] A computer-readable storage medium may be a tangible device capable of holding and storing instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), static random access memories (SRAMs), portable compact disk read-only memories (CD-ROMs), digital versatile disks (DVDs), memory sticks, floppy disks, mechanically encoded devices such as punch cards or ridge structures in grooves with instructions recorded thereon, and any suitable combinations of the above. As used herein, a computer-readable storage medium should not be construed as being a transitory signal per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., light pulses passing through a fiber optic cable), or electrical signals transmitted over wires.

[0058] The computer-readable program instructions described herein can be downloaded to each computing / processing device from a computer-readable storage medium or via an external computer or external storage device over a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may comprise copper transmission cables, optical transmission fiber, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing / processing device.

[0059] The computer readable program instructions for carrying out the operations of the present disclosure may be either source or object code written in any combination of one or more programming languages, including assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, or object oriented programming languages ​​such as Smalltalk, C++, and traditional procedural programming languages ​​such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or wide area network (WAN), or a connection may be made to an external computer (e.g., via the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may execute computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry to perform aspects of the present disclosure.

[0060] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It can be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0061] These computer readable program instructions may be provided to a processor of a general purpose computer, a special purpose computer, or other programmable data processing apparatus to manufacture a machine such that the instructions, executed by the processor of the computer or other programmable data processing apparatus, create means for performing the functions / operations specified in one or more blocks of the flowcharts and / or block diagrams. These computer readable program instructions may also be stored in a computer readable storage medium capable of directing a computer, programmable data processing apparatus, and / or other device to function in a particular manner, such that the computer readable storage medium having stored thereon instructions comprises an article including instructions implementing aspects of the functions / operations specified in one or more blocks of the flowcharts and / or block diagrams.

[0062] Furthermore, the computer readable program instructions may be loaded onto a computer, other programmable apparatus, or other device to cause the computer, other programmable apparatus, or other device to perform a series of operational steps to generate a computer-implemented process, such that the instructions, executing on the computer, other programmable apparatus, or other device, perform the functions / acts specified in one or more blocks of the flowcharts and / or block diagrams.

[0063] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or part of instructions that includes one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions described in the blocks may be performed out of the order described in the figures. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may sometimes be executed in the reverse order, depending on the functionality involved. It should also be noted that each block of the block diagrams and / or flowchart diagrams, and combinations of blocks in the block diagrams and / or flowchart diagrams, may be implemented by a dedicated hardware-based system that executes the specified functions or operations or executes a combination of dedicated hardware and computer instructions.

[0064] The description of various embodiments of the present disclosure is presented for illustrative purposes, but is not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terms used in this specification are selected to best explain the principles of the embodiments, practical applications, or technical improvements to the technology found in the market, or to enable those skilled in the art to understand the embodiments disclosed herein.

[0065] example Example 1: Method Data Sources and Study Population: The CA cohort consisted of 2541 patients seen at Mayo Clinic between January 1, 2000 and May 31, 2019, identified from the institutional amyloid database. Inclusion criteria included a diagnosis of ATTR or AL with cardiac involvement and a digital 12-lead ECG within 180 days of diagnosis. A diagnosis of AL required a biopsy (cardiac or noncardiac) with positive staining for amyloid and typing by immunohistochemistry, immunofluorescence, or laser microdissection mass spectrometry. ATTR was diagnosed based on either a tissue diagnosis or abnormalities detected on technetium Tc 99m pyrophosphate scintigraphy in the absence of evidence of AL amyloidosis. For AL and ATTRv, cardiac involvement was defined as either an interventricular septal thickness of more than 12 mm, a troponin T level of more than 0.03 ng / L or an N-terminal pro-B-type natriuretic peptide level of more than 332 pg / mL, or any cardiac symptoms not attributable to other cardiac conditions. All patients with ATTRwt were considered to have cardiac involvement. Exclusion criteria included normal findings on endomyocardial biopsy, transthoracic echocardiography interpreted as negative for cardiac amyloid, and ventricular pacing at the time of ECG acquisition. Controls were selected from Mayo Clinic clinics who were not present in the amyloid cohort and had at least one digital 12-lead ECG performed within 180 days of the transthoracic echocardiography. Controls with echocardiograms interpreted as consistent with amyloidosis or infiltrative cardiomyopathy were excluded. Patients with previous cardiovascular surgery, poorly reported ECGs and echocardiograms, ECGs with invalid waveforms or ventricular pacing, and left ventricular ejection fractions less than 50% were excluded from the control group. Cases and controls were matched using a nearest neighbor algorithm for age and sex in a 1:1 ratio. ECGs in both the amyloid and control groups were digital, standard, 10-second, 12-lead ECGs obtained in the supine position. In addition to the 12-lead ECG, single-lead and 6-lead ECG recordings were obtained.

[0066] Model Development: Three datasets - training (n=2997; 60%), validation (n1 / 4999; 20%) and testing (n=999; 20%) - were prepared by assigning cases and controls by outcome-stratified random sampling. These proportions were chosen to ensure an adequate number of cases in the subsets, with each patient uniquely assigned to a single group.

[0067] The model architecture used is shown in Figure 10, where the input consists of raw ECG waveform data and a binary output predicting the presence or absence of CA. It should be noted that only raw data was used, no manually extracted features were used.

[0068] We implemented a convolutional neural network (CNN) using the Keras Framework with Tensorflow (Google, Mountain View, CA) backend and Python. Although CNNs are primarily applied to images, we tweaked the network architecture to have spatial and temporal feature extraction layers. The network works by adjusting the weights of the convolutional filters during training to extract meaningful and relevant features in an unsupervised manner. We built the network using stacked blocks of convolution, max pooling, and batch normalization, followed by a nonlinear activation function for each block. After the first group of blocks extracted the temporal features, another spatial block was used to fuse the data from all the leads, and the extracted features were then used in a fully connected network. Of note, the only data input for training was the raw digital 12-lead ECG signal and the associated classification of each individual. Figure 10 summarizes the network architecture of the exemplary network. The hyperparameters (batch size, initial learning rate, number of neurons in the fully connected layer, and number of convolutional layers) were varied during training to obtain the best model based on the validation set.

[0069] Initial experiments were performed using transfer learning, updating the weights of the existing network very slowly or freezing them completely, but the best performance was found by retraining from scratch. Training consisted of 30 epochs run with a batch size of 32 and a learning rate of 0.001. A categorical cross-entropy loss was used with one-hot coded indicators of either ATTR or AL as the positive case and noamyloid as the negative case. The optimizer was Adam and the final layer activation was softmax. 16 Alternative combinations of batch size and learning rate were systematically evaluated using the validation set with the one reported herein with the highest performance. During training, after each epoch, the network was evaluated on the validation set and the area under the receiver operating characteristic curve (AUC) was recorded. At the end of 30 epochs, the model with the highest validation set AUC was selected as optimal and evaluated on the test set.

[0070] The output probability for classifying a given patient was compared to a threshold that could be adjusted to different clinical scenarios, for example emphasizing sensitivity over specificity. Our primary analysis adjusted the threshold towards a balance between the two by selecting the point on the receiver operating characteristic curve of the validation set with the highest Youden index and designating the corresponding threshold as optimal. Single-lead and six-lead models were developed using similar architectures, but with adapted layer shapes.

[0071] Primary Analysis: AI-enabled ECG to detect CA: The effectiveness of the model in detecting CA was evaluated using the closest 12-lead ECG within 180 days from the date of diagnosis. Performance in distinguishing patients with CA from non-amyloid controls was assessed by use of receiver operating characteristic curves, and sensitivity and specificity were evaluated at two thresholds, one favoring higher sensitivity and the other determined by maximizing the Youden index. Secondary analyses were performed using single-lead and 6-lead tracings.

[0072] Subgroup analysis: Model performance was evaluated in subsets of patients by age, sex, and amyloid subtype (AL, ATTRwt, and ATTRv), as well as in patients who met electrocardiographic criteria for low voltage, LVH, or "normal" ECG. Additionally, baseline comorbidities and 12-lead ECG findings for CA patients and controls were reported. Categorical variables are reported as absolute numbers and percentages, and continuous variables are reported as means with standard deviations. Categorical variables were compared with the c2 test, and continuous variables were compared using the Student's t test. Data were analyzed using the R software package v3.6.2, 18 Statistical analysis was performed using and model development was performed using TensorFlow 2.1.0 and Python 3.7.7. AP values ​​less than 0.005 were considered significant.

[0073] Temporal analysis: A temporal analysis was performed to evaluate the model's ability to detect CA before the clinical diagnosis date. For each patient with CA, all available ECGs recorded 6 months to 5 years prior to diagnosis were retrieved. The model output of amyloid probability was recorded for each follow-up, along with the time from acquisition to the date of diagnosis. The distribution of amyloid probability over time was assessed by Violin plot. For patients with multiple ECGs within a given time bin, the median predicted amyloid probability score was used.

[0074] Example 2: Results Study population: Baseline characteristics, comorbidities, and 12-lead ECG findings at diagnosis are outlined in Tables 1-4. ATTRwt patients were mostly male and over 60 years of age. CA patients more commonly had 12-lead ECG abnormalities compared with controls in all categories except left bundle branch block and QT interval prolongation. Atrial arrhythmias and conduction system disease were more common in ATTRwt than AL, whereas low voltage ECG was more common in AL. Criteria for LVH were present in nearly 20% of ATTRwt patients. ECG characteristics between the training, test, and validation sets were comparable (Table 4).

[0075] [Table 1]

[0076] [Table 2]

[0077] [Table 3]

[0078] [Table 4]

[0079] Model performance: The test set receiver operating characteristic curve for the 12-inductor network is shown in Figure 5. The AUC of 0.91 (95% CI, 0.90-0.93) indicates excellent performance. Using an optimal probability threshold of 0.485, the sensitivity was 0.84 and the specificity was 0.85. In this 1:1 matched population, the positive predictive value was 0.86 and the negative predictive value was 0.84. The calibration curve (Figure 6) shows excellent linearity. Table 5 shows examples of performance at different probability thresholds and prevalences, including the balanced threshold as well as thresholds selected to make the test highly specific. Model predictions of CA probability in the control group are shown as violin plots in Figure 7.

[0080] [Table 5]

[0081] Single-lead and six-lead model performance was evaluated. V5 was the best single-lead model with an AUC of 0.86 and accuracy of 0.78, and other single-leads performed similarly. The top six-lead (1, 2, 3, aVR, aVL, aVF) model had an AUC of 0.90 and accuracy of 0.85.

[0082] Subgroup analysis: Model performance was similar across gender and age, and amyloid subtype (Table 6). Specificity was lower in those with low voltage or LVH and normal ECG interpretations, with lower sensitivity but higher specificity.

[0083] [Table 6]

[0084] Temporal analysis: ECG was available within a range of 5 years to 6 months prior to clinical diagnosis in 396 patients with CA. Among patients with CA and pre-diagnosis ECG studies, the AI ​​model successfully predicted the presence of CA >6 months prior to clinical diagnosis in 234 (59%), identifying 46% of AL and 70% of ATTRwt. Figures 8A-B show the distribution of amyloid probability over time prior to diagnosis in 6-month increments of AL and ATTRwt. For AL patients, roughly half of the patients met the 0.485 threshold 12 months prior to diagnosis, and for ATTR, the score was above the median for most patients 48 months prior to diagnosis. An example of amyloid detection by the AI-ECG model 28 months prior to clinical diagnosis of AL amyloidosis is outlined in Figures 9A-B.

[0085] Example 3: Discussion Amyloid heart disease is associated with high morbidity and mortality, and diagnosis is often delayed, resulting in poor outcomes. Given the advent of effective treatments, there is a pressing clinical need for an easily deployable and scalable test for early detection. 12-lead ECGs are universal and inexpensive, making them ideally suited to be transformed into a tool that facilitates early diagnosis of CA by the addition of AL. There are four key findings in this study. First, CA can be accurately predicted with an AI 12-lead ECG alone. Second, the model performed well across amyloid subtypes. Third, in the subset of patients with available serial ECGs, the model demonstrated that CA was nearly 60% ahead of clinical diagnosis. Fourth, the tool can be adapted to 6-lead and single-lead ECG acquisition, enabling point-of-care screening with smartphone-compatible electrodes.

[0086] Cardiac amyloidosis has been considered a rare disease, but studies suggest that it is underdiagnosed. The true prevalence of CA is unknown, but it is an area of ​​active investigation. While light chain (AL) amyloidosis still appears to be relatively rare, recent evidence suggests that ATTRwt is probably not rare at all, being present in over 15% of patients undergoing transcatheter aortic valve replacement, as well as 13% of patients with heart failure with preserved ejection fraction. The V122I TTR mutation is present in 3%-4% of Africans in the United States, resulting in a population of approximately 1.5 million individuals at risk for CA. Unfortunately, patients with CA are often misdiagnosed as having hypertrophic cardiomyopathy, hypertensive heart disease, or other causes of heart failure, leading to delayed or inappropriate treatment.

[0087] The pathophysiological mechanisms of CA are complex and include not only myocardial infiltration but also direct toxic effects on the heart, therefore the term toxic infiltrative cardiomyopathy is more accurate. In addition to extracellular infiltration of the myocardium and replacement of cardiac tissue with electrically inactive amyloid deposits, circulating light chains in AL amyloidosis cause myocyte dysfunction, and similar toxic effects may occur in ATTR amyloidosis. Myocyte hypertrophy has been reported in ATTRwt amyloidosis, which may explain the electrocardiographic criteria findings of LVH in almost 20% of these patients.

[0088] Because the AI ​​model can detect multiple simultaneous ECG features, it has the potential to detect these complex physiological changes early in the disease course. In other conditions, such as ischemia and left ventricular dysfunction, electrical changes appear before structural changes are detected by echocardiography and well before the onset of symptoms. The superior performance of the models in both AL and ATTR, despite significant differences in baseline 12-lead characteristics, suggests that the AI ​​model can "see" physiological changes specific to amyloid that are not recognized by traditional ECG interpretation.

[0089] Advances in noninvasive imaging, including echocardiography and cardiac magnetic resonance imaging, have greatly improved the diagnosis of CA. Echocardiography is a powerful tool to suggest the diagnosis, but differentiation from other causes of increased wall thickness remains difficult. Variability in the extent and distribution of CA can result in nonclassical echocardiographic and cardiac magnetic resonance imaging findings, leading to delayed diagnosis. Some patients with AL amyloidosis have normal or slightly increased wall thickness despite rapidly progressive disease, and establishing the diagnosis several months in advance may allow valuable time to respond to treatment.

[0090] Therefore, the key to the diagnosis of CA is a high index of clinical suspicion. Most patients ultimately diagnosed with CA had an ECG performed at some point in the diagnostic process. Diagnostic algorithms are readily available, so the key for clinicians is to consider the diagnosis. Apart from developing a definitive diagnostic test, methods are disclosed herein that utilize data already obtained by electrocardiography. AL and ATTR were included because early diagnosis is important in both, typing is performed by other techniques, and the inclusion of ATTR expands the utility of this model to "less rare" forms of amyloid. The AI ​​model is intended to maximize the diagnostic yield of the ECG, including suggesting diagnoses that have not yet been considered. The work disclosed herein found that an AI model using only ECG can accurately predict CA without the need for other clinical or imaging variables. The accuracy of single-lead and six-lead acquisitions allows for implementation by mobile devices.

[0091] We evaluated model performance across different amyloid subtypes and demonstrated for the first time the ability of an AI-ECG tool to detect amyloid heart disease prior to clinical diagnosis. The 12-lead model was extended to single-lead and 6-lead ECG acquisitions, which have not been previously reported.

[0092] Losing AI-enhanced models of ECG analysis for the detection of low ejection fraction, atrial fibrillation, and hypertrophic cardiomyopathy led to the creation of an "AI-ECG dashboard" for use in clinical practice. This tool can be accessed via the electronic medical record and easily integrated into practice. With further validation, the incorporation of an AI-ECG model for the detection of CA will provide an important new tool to facilitate timely diagnosis and improved outcomes.

[0093] The use of the AI-ECG model may prove useful in determining prognosis, risk of sudden cardiac death, and response to therapy. The model may allow early detection of cardiac involvement in ATTRv carriers or in patients with ATTR detected in non-cardiac tissues such as the carpal tunnel and spinal ligaments. Although ATTRwt amyloidosis has been reported to have a male predominance of over 90% in most studies to date, women do develop the condition and may be underdiagnosed. The AI-ECG model may be able to suggest a diagnosis of ATTRwt amyloidosis in women without classic echocardiographic findings. The ability to use single-lead and six-lead acquisitions holds promise for simple, cost-effective, global screening of at-risk populations, especially those with limited access to advanced cardiac imaging.

[0094] Cardiac amyloidosis can occur well before clinical diagnosis, resulting in electrocardiogram changes that are detected by the application of AI to standard ECG, a universal and inexpensive test. Despite increased awareness and improved imaging techniques, delayed diagnosis of CA still has dire consequences. The use of this AI-ECG model to detect CA may facilitate earlier diagnosis and initiation of potentially life-saving therapy.

[0095] References 1.Muchtar E, Gertz MA, Kumar SK et al. Improved outcomes in newly diagnosed AL amyloidosis from 2000 to 2014: Cracking the glass ceiling of premature mortality. Blood. 2017;129(15):2111-2119. 2. Lousada I, Comenzo RL, Landau H, Guthrie S, Merlini G. Light chain amyloidosis: a patient experience survey from the Amyloidosis Research Consortium. Adv Ther. 2015;32(10):920-928. 3. Ruber FL, Grogan M, Hanna M, Kelly JW, Maurer MS. Trans-thyretin amyloid cardiomyopathy: a JACC state-of-the-art review. J Am Coll Cardiol. 2019;73(22):2872-2891. 4. Alexander KM, Evangelisti A, Witteles RM. Diagnosis and treatment of cardiac amyloidosis associated with plasma cell dyscrasias. Cardiol Clin. 2019;37(4):487-495. 5. Maurer MS, Schwartz JH, Gundapaneni B et al. Tafamidis treatment for patients with transthyretin amyloid cardiomyopathy. N Engl J Med.2018;379(11):1007-1016. 6. Falk RH, Alexander KM, Liao R, Dorbala S. AL (light chain) cardiomyopathy: a review of diagnosis and treatment. J Am Coll Cardiol. 2016;68(12):1323-1341. 7. Gertz MA, Dispenzieri A, Sher T. Pathophysiology and treatment of cardiac amyloidosis. Nat Rev Cardiol. 2015;12(2):91-102. 8. Grogan M, Scott CG, Kyle RA et al. Natural history of wild-type transthyretin cardiac amyloidosis and risk stratification using a new staging system [correction published in J Am Coll Cardiol. 2017;69(23):2882]. J Am Coll Cardiol. 2016;68(10):1014-1020. 9. Gillmore D, Maurer MS, Falk RH et al. Non-biopsy diagnosis of cardiac transthyretin amyloidosis. 2016;133(24):2404-2412. 10. Sperry BW, Ikram A, Hachamovitch R et al. Efficacy of chemotherapy for light-chain amyloidosis in patients with symptomatic heart failure. J Am Coll Cardiol. 2016;67(25):2941-2948. 11. Murtagh B, Hammill SC, Gertz MA, Kyle RA, Tajik AJ, Grogan M. Electrocardiographic findings in primary systemic amyloidosis and biopsy-proven cardiac involvement. Am J Car-diol. 2005;95(4):535-537. 12.Ko WY,Siontis KC,Attia ZI et al.Detection of hypertrophic cardiomyopathy using convolutional neural network-enabled electrocardiograms.J Am Coll Cardiol.2020;75(7):722-733. 13. Vrana JA. Classification of amyloidosis by laser microdissection and mass spectrometry-based proteomic analysis of clinical biopsy specimens. Blood. 2009;114(24):4957-4959. 14. Dorbala S, Ando Y, Bokhari S et al. ASNC / AHA / ASE / EANM / HFSA / ISA / SCMR / SNMMI Expert consensus recommendations for multimodality imaging in cardiac amyloidosis: Part 2 of 2d diagnostic criteria and appropriate use. J Card Fail. 2019;25(11):854-865. 15. Attia ZI, Friedman PA, Noseworthy PA et al. Estimating age and gender from standard 12-lead ECGs using artificial intelligence. Circ Arrhythm Electrophysiol. 2019;12(9):e007284. 16. Goodfellow I, Bengio Y, Courville A. Deep Learning. Cambridge, MA: The MIT Press; 2016. 17. Youden WJ. An index for evaluating diagnostic tests. Cancer. 1950;3(1):32-35. 18. R Core Team. R: a language and environment for statistical computing, version 3.5.3. Vienna, Austria: R Foundation for Statistical Computing; 2019. 19. Gurwitz JH, Maurer MS. Tafamidisda: a cost-effective treatment for a not-so-rare condition. JAMA Cardiol. 2020;5(3):247-248. 20.Castano A,Narotsky DL,Hamid N et al. Transthyretin cardiac amyloidosis and its predictors in elderly patients with severe aortic stenosis undergoing transcatheter aortic valve replacement.Eur Heart J.2017;38(38):2879-2887. 21. Gonzalez-Lopez E, Gallego-Delgado M, Guzzo-Merello G et al. Wild-type transthyretin amyloidosis as a cause of heart failure with preserved ejection fraction. Eur Heart J. 2015;36(38):2585-2594. 22. Buxbaum JN, Ruberg FL. Transthyretin V122I (pV142I)* cardiac amyloidosis: an under-recognized age-related autosomal dominant heart disease in older African Americans. Genet Med. 2017;19(7):733-742. 23. Rapezzi C, Merlini G, Quarta CC. Systemic cardiac amyloidoses: disease profile and clinical course of the three major types. Circulation. 2009;120(13):1203-1212. 24. Maleszewski JJ. Cardiac amyloidosis: pathology, nomenclature and typing. Cardiovascul Vascular Pathways. 2015;24(6):343-350. 25.Fealey ME, Edwards WD, Buadi FK, Syed IS, Grogan M. Echocardiographic features of cardiac amyloidosis presenting as endomyocardial disease in a 54-year-old man. J Cardiol.2009;54(1):162-166. 26. Suresh R. Advanced cardiac amyloidosis associated with normal interventricular septal thickness: a rare manifestation of infiltrative cardiomyopathy. J Am Soc Echocardiogram 2014;27(4):440-447. 27. Tison GH, Zhang J, Delling FN, Deo RC. Automated, preconfigurable patient ECG profiles for disease detection, tracking, and discovery. Circulation Cardiovasc Qual Outcomes. 2019;12(9):e005289.

[0096] All publications (including patents, patent applications, and sequence accession numbers referred to herein) are herein incorporated by reference in their entirety as if each individual publication was specifically and individually indicated to be incorporated by reference. In case of conflict, the present application, including any definitions herein, will control.

[0097] Those skilled in the art will recognize, or be able to ascertain using no more than routine experimentation, many equivalents to the specific embodiments of the invention described herein which equivalents are intended to be encompassed by the following claims.

Claims

**Claim 1** Receiving voltage-time data of a subject, wherein the voltage-time data includes voltage data of a plurality of leads of an electrocardiograph; Generating a feature vector from the voltage-time data; Providing the feature vector to a pre-trained learning system; Receiving, from the pre-trained learning system, an indication of the presence or absence of cardiac amyloidosis (CA) in the subject; A method comprising the above steps. **Claim 2** The method according to claim 1, wherein generating the feature vector includes generating a spectrogram based on the voltage data of the plurality of leads. **Claim 3** The method according to claim 1, wherein generating the feature vector includes grouping the voltage data of the plurality of leads into a plurality of subsets. **Claim 4** The method further includes receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector. The method according to claim 1. **Claim 5** The method further includes receiving genomic information of the subject, and generating the feature vector includes adding the genomic information to the feature vector. The method according to claim 1. **Claim 6** The method according to claim 1, wherein the learning system comprises a convolutional neural network. **Claim 7** The method according to claim 6, wherein the convolutional neural network includes at least one residual connection. **Claim 8** The method according to claim 1, wherein the voltage-time data of the subject is received from an electrocardiograph. **Claim 9** The method according to claim 1, wherein the voltage-time data of the subject is received from an electronic medical record. **Claim 10** The method according to claim 1 further includes providing the indication to an electronic health record system for storage in a health record associated with the subject. The method according to claim 1. **Claim 11** The method according to claim 1 further includes providing the indication to a computing node for display to a user. The method according to claim 1. **Claim 12** The method according to claim 1, wherein the feature vector includes a matrix having a plurality of rows and a plurality of columns, the plurality of rows corresponding to a time dimension, and the plurality of columns corresponding to a spatial dimension. **Claim 13** The method according to claim 12, wherein each of the plurality of rows corresponds to one of the plurality of leads, and each of the plurality of columns corresponds to a timestamp. **Claim 14** The method according to claim 12, wherein the time dimension has a resolution of 500 Hz.

15. The method according to claim 6, wherein the convolutional neural network comprises at least nine convolutional blocks and two fully connected blocks.

16. An electrocardiograph comprising a plurality of leads, A computing node comprising a computer-readable storage medium embodied with program instructions, the program instructions being executable by a processor of the computing node, whereby the processor Receiving voltage-time data of a subject from an echocardiograph, the voltage-time data including the voltage data of the plurality of leads; Generating a feature vector from the voltage-time data; Providing the feature vector to a pre-trained learning system; Receiving an indication of the presence or absence of CA in the subject from the pre-trained learning system; A computing node that executes a method including the above steps; A system comprising the above components.

17. The system according to claim 16, wherein generating the feature vector includes generating a spectrogram based on the voltage data of the plurality of leads.

18. The system according to claim 16, wherein generating the feature vector includes grouping the voltage data of the plurality of leads into a plurality of subsets.

19. Further comprising receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector. The system according to claim 16.

20. Further comprising receiving genomic information of the subject, and generating the feature vector includes adding the genomic information to the feature vector. The system according to claim 16.

21. The system according to claim 16, wherein the learning system comprises a convolutional neural network.

22. The system according to claim 21, wherein the convolutional neural network includes at least one residual connection.

23. The system according to claim 16, wherein the voltage-time data of the subject is received from an electrocardiograph.

24. The system according to claim 16, wherein the voltage-time data of the subject is received from an electronic medical record.

25. Providing the display to an electronic health record system for storage in a health record associated with the subject The system according to claim 16, further comprising this.

26. Providing the display to a computing node for display to a user The system according to claim 16, further comprising this.

27. The system according to claim 16, wherein the feature vector includes a matrix having a plurality of rows and a plurality of columns, the plurality of rows corresponding to a time dimension, and the plurality of columns corresponding to a spatial dimension.

28. The system according to claim 27, wherein each of the plurality of columns corresponds to one of the plurality of leads, and each of the plurality of columns corresponds to a timestamp.

29. The system according to claim 27, wherein the time dimension has a resolution of 500 Hz.

30. The system according to claim 21, wherein the convolutional neural network comprises at least nine convolutional blocks and two fully connected blocks.

31. A computer program product for detecting cardiac amyloidosis (CA), comprising a computer-readable storage medium embodied with program instructions, the program instructions being executable by a processor, whereby the processor Receiving voltage-time data of a subject from an echocardiogram, the voltage-time data including voltage data of the plurality of leads; Generating a feature vector from the voltage-time data; Providing the feature vector to a pre-trained learning system; Receiving an indication of the presence or absence of CA in the subject from the pre-trained learning system; A computer program product that executes a method including this.

32. The computer program product according to claim 31, wherein generating the feature vector includes generating a spectrogram based on the voltage data of the plurality of leads.

33. The computer program product according to claim 31, wherein generating the feature vector includes grouping the voltage data of the plurality of leads into a plurality of subsets.

34. Further comprising receiving demographic information of the subject, and generating the feature vector includes adding the demographic information to the feature vector, The computer program product according to claim 31.

35. Further comprising receiving genomic information of the subject, and generating the feature vector includes adding the genomic information to the feature vector. The computer program product according to claim 31.

36. The computer program product according to claim 31, wherein the learning system comprises a convolutional neural network.

37. The computer program product according to claim 36, wherein the convolutional neural network includes at least one residual connection.

38. The computer program product according to claim 31, wherein the voltage-time data of the subject is received from an electrocardiograph.

39. The computer program product according to claim 31, wherein the voltage-time data of the subject is received from an electronic medical record.

40. Further comprising providing the display to an electronic health record system for storage in a health record associated with the subject. The computer program product according to claim 31.

41. Further comprising providing the display to a computing node for display to a user. The computer program product according to claim 31.

42. The feature vector includes a matrix having a plurality of rows and a plurality of columns, the plurality of rows corresponding to a time dimension, and the plurality of columns corresponding to a spatial dimension. The computer program product according to claim 31.

43. Each of the plurality of columns corresponds to one of the plurality of leads, and each of the plurality of columns corresponds to a timestamp. The computer program product according to claim 42.

44. The computer program product according to claim 42, wherein the time dimension has a resolution of 500 Hz.

45. The computer program product according to claim 36, wherein the convolutional neural network comprises at least nine convolutional blocks and two fully connected blocks.

Images