System and method for distinguishing non-failing from failing cardiac fibroblasts

Abstract

A system and method for distinguishing non-failing from failing cardiac fibroblasts (CFs) using advanced imaging techniques and machine learning (ML). The method involves staining CFs with fluorescent dyes to highlight key cellular components, such as nuclei and actin fibers. Images are analyzed to extract morphological features, which are then processed and normalized. A supervised ML system, trained on labeled CF data, classifies CFs based on phenotypic differences. The system enables accurate diagnosis of heart failure (HF), facilitates screening for therapeutic efficacy, and identifies antifibrotic agents by evaluating changes in CF phenotypes pre- and post-treatment.

Description

[0001] SYSTEM AND METHOD FOR DISTINGUISHING NONFAILING FROM FAILING CARDIAC FIBROBLASTS

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] This nonprovisional application claims priority to provisional application No. 63 / 735,399, entitled “A System and Method For Distinguishing Non-failing From Failing Cardiac Fibroblasts,” filed December 18, 2024 by the same inventor(s) and to provisional application No. 63 / 764,719, entitled “A System and Method For Distinguishing Non-failing From Failing Cardiac Fibroblasts,” filed February 28, 2025 by the same inventor(s).

[0004] GOVERNMENT SUPPORT

[0005] This invention was made with government support under Grant Nos. HL 166708, HL 150225, HL 127240, HL 147558, and HL 116848, all awarded by the National Institutes of Health (NIH). This invention was also made with foundational support under Grant No. 24SCA1255857, awarded by the American Heart Association (AHA). The government has certain rights in the invention.

[0006] BACKGROUND OF THE INVENTION

[0007] Heart failure, the final clinical manifestation of numerous forms of cardiovascular disease, is a devastating syndrome with poor prognosis characterized by chamber remodeling and reduced ventricular compliance. It remains the leading cause of death in the United States accounting for nearly one million deaths annually and imposes a significant economic burden with the cost of cardiovascular diseases projected to approach $1 trillion by 2030.

[0008] In response to injury, cardiac fibroblasts (CFs) undergo a cell state transition to become activated fibroblasts, which produce excessive amounts of extracellular matrix proteins and thereby contribute to pathological fibrotic remodeling of the heart1. Activated CFs are typically defined by imaging-based detection of a-smooth muscle actin (a-SMA)- containing stress fibers. a-SMA staining has been used to screen for inhibitors of agonist-dependent CF activation2. Furthermore, on the basis of elevated a-SMA levels, it has been proposed that, compared to non-failing controls, CFs derived from failing human hearts maintain an activated state in culture, which is indicative of epigenetic memory2 3.

[0009] Unfortunately, interventions targeting the CF population do not yet exist in the clinical realm despite the profound contributions of fibrotic remodeling to the development of cardiac dysfunction and heart failure, and the discovery of novel antifibrotic strategies is anticipated to be a promising avenue for therapeutic intervention. Historically, drug discovery efforts targeting fibroblast activation have generally relied upon only a handful of validated biomarkers of activated myofibroblasts in culture, perhaps most notably the formation of a-SMA stress fibers, which play an important role in the contractile function of fibroblasts in the fibrotic heart. Accordingly, systems and methods for detecting and analyzing cellular changes in cardiac tissue to distinguish between healthy and pathological states of CFs and facilitate the diagnosis and treatment of heart disease and fibrosis-related conditions are needed. However, in view of the art considered as a whole at the time the present invention was made, it was not obvious to those of ordinary skill in the field of this invention how the shortcomings of the prior art could be overcome.

[0010] All referenced publications are incorporated herein by reference in their entirety. Furthermore, where a definition or use of a term in a reference, which is incorporated by reference herein, is inconsistent or contrary to the definition of that term provided herein, the definition of that term provided herein applies and the definition of that term in the reference does not apply.

[0011] While certain aspects of conventional technologies have been discussed to facilitate disclosure of the invention, Applicants in no way disclaim these technical aspects, and it is contemplated that the claimed invention may encompass one or more of the conventional technical aspects discussed herein.

[0012] The present invention may address one or more of the problems and deficiencies of the prior art discussed above. However, it is contemplated that the invention may prove useful in addressing other problems and deficiencies in a number of technical areas. Therefore, the claimed invention should not necessarily be construed as limited to addressing any of the particular problems or deficiencies discussed herein.

[0013] In this specification, where a document, act or item of knowledge is referred to or discussed, this reference or discussion is not an admission that the document, act or item of knowledge or any combination thereof was at the priority date, publicly available, known to the public, part of common general knowledge, or otherwise constitutes prior art under the applicable statutory provisions; or is known to be relevant to an attempt to solve any problem with which this specification is concerned.

[0014] BRIEF SUMMARY

[0015] Aspects of the present disclosure generally relate to a method, apparatus, system, computer program product, non-transitory computer-readable medium, and / or processing system as substantially described herein and as illustrated in the accompanying drawings and specification. In some embodiments, the aspects may include combinations or sub combinations of the elements described herein, as would be understood by one of ordinary skill in the art.

[0016] Some implementations of the present invention relate to a method for analyzing cardiac fibroblasts. In some implementations, the method includes acquiring one or more multi-channel fluorescence images of stained cardiac fibroblasts; processing the images by performing illumination correction on each fluorescence channel, segmenting nuclei, cytoplasm, and whole-cell regions, extracting a plurality of morphological features, filtering segmentation errors or debris, curating single-cell feature data, and normalizing the curated data to generate singlecell profiles; providing the single-cell profiles to a machine-learning system configured to classify cardiac fibroblasts; and classifying the cardiac fibroblasts as failing or non-failing based on the input data.

[0017] Additional features of the method may include staining cardiac fibroblasts with F-actin stains; extracting morphological features from nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus; extracting features of at least nuclei and cytoplasm; extracting highly informative features including the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm; organizing curated features into tabular formats; training the machine-learning system using normalized single-cell profiles and ground-truth labels; diagnosing heart failure or predicting likelihood of heart failure from the classification; assessing therapeutic efficacy using pre- and post-treatment images; and administering a therapeutic agent if the cardiac fibroblasts are classified as failing.

[0018] Some implementations of the present invention relate to a method for predicting heart failure in a subject. In some implementations, the method includes processing multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to extract and normalize morphological features; providing the resulting single-cell profiles to a machine-learning system trained using training data from subjects clinically characterized as having or not having heart failure; and predicting, by the machine-learning system, whether the subject has heart failure based on the extracted morphological features.

[0019] Additional features of the method may include training the machine-learning system using single-cell profiles associated with clinical labels indicating heart failure status; adjusting model parameters based on prediction errors during training; using key predictive features including Hoechst edge intensities and actin edge intensities; assigning weighting coefficients to morphological features based on predictive importance; producing probability scores or risk indices; staining CFs with F-actin; and extracting features from nuclei, cytoplasm, and other cellular organelles.

[0020] Some implementations of the present invention relate to a method for diagnosing heart failure in a subject. In some implementations, the method includes processing multi-channel fluorescence images of cardiac fibroblasts to extract and normalize morphological features; providing the extracted features to a machine-learning system trained using clinical heart-failure labels; and determining whether the subject has heart failure based on the machine-learning system’s output.

[0021] Additional features of the method may include training the machine-learning system using clinically labeled cardiac fibroblasts; assigning feature-weighting coefficients based on diagnostic importance; generating predicted clinical labels during training and adjusting model parameters based on evaluation metrics; extracting morphological features from multiple cellular compartments; using diagnostically informative features such as Hoechst edge intensities and actin edge intensities; and performing the analysis on CFs stained with F-actin.

[0022] Some implementations of the present invention relate to a method for screening the efficacy of a therapeutic agent. In some implementations, the method includes acquiring first multi-channel fluorescence images of cardiac fibroblasts before exposure to the therapeutic agent and second images after exposure; processing the images into normalized single-cell profiles; comparing morphological features between timepoints; and determining the efficacy of the therapeutic agent based on changes in one or more morphological features.

[0023] Additional features of the method may include determining whether morphological features move toward values characteristic of non-failing cardiac fibroblasts; comparing features extracted from nuclei and cytoplasm; determining whether features increase, decrease, or cross threshold boundaries; comparing Hoechst edge intensities and actin edge intensities; providing pre- and post-treatment profiles to a machine-learning system to compute phenotypic-change metrics; determining efficacy when treated profiles shift toward non-failing profiles; applying rule-based or threshold-based criteria; and acquiring images from additional intermediate or later timepoints.

[0024] Some implementations of the present invention relate to a method of treating a subject. In some implementations, the method includes acquiring multi-channel fluorescence images of cardiac fibroblasts obtained from the subject; processing the images to extract morphological features; classifying the cardiac fibroblasts as failing or non-failing based on the morphological features; and administering a therapeutic agent to the subject when the cardiac fibroblasts are classified as failing.

[0025] Additional features of the method may include providing single-cell profiles to a machinelearning system trained with heart-failure clinical labels; determining whether morphological features deviate toward failing values; evaluating diagnostic features including Hoechst edge intensities and actin edge intensities; administering treatment indicated for heart failure, fibrosis, or cardiac remodeling; analyzing post-treatment CFs to determine treatment response; adjusting therapy based on post-treatment CF morphology; applying thresholds to determine treatment triggers; and extracting features from multiple cellular compartments including nuclei, cytoplasm, mitochondria, ER, Golgi, and plasma membrane.

[0026] Some implementations of the present invention relate to a system for analyzing cardiac fibroblasts. In some implementations, the system includes one or more processors and memory storing instructions that cause the system to acquire multi-channel fluorescence images of stained cardiac fibroblasts; perform illumination correction; segment nuclei, cytoplasm, and whole-cell regions; extract morphological features; filter segmentation errors; curate and normalize feature data to generate single-cell profiles; and provide the profiles to a machinelearning system configured to classify cardiac fibroblasts.

[0027] Additional features of the system may include staining CFs with F-actin; extracting features from multiple cellular compartments; extracting key edge-intensity and actin-intensity features; generating tabular single-cell datasets; training the machine-learning system using CFs labeled as failing or non-failing; diagnosing heart failure; predicting a likelihood of heart failure; comparing pre- and post-treatment morphological features; and recommending therapeutic interventions based on classification.

[0028] Some implementations of the present invention relate to a system for diagnosing heart failure in a subject. In some implementations, the system includes processors and memory storing instructions to process multi-channel fluorescence images of cardiac fibroblasts, generate normalized single-cell profiles, and provide the profiles to a machine-learning system trained using clinically labeled heart-failure data to diagnose whether the subject has heart failure.

[0029] Additional features of the system may include training the machine-learning system using clinical labels; weighting morphological features based on diagnostic relevance; evaluating model performance during training; extracting features from nuclei, cytoplasm, mitochondria, ER, Golgi, and plasma membrane; identifying diagnostically relevant features including Hoechst and actin edge-intensity features; and analyzing CFs stained with F-actin.

[0030] Some implementations of the present invention relate to a system for screening the efficacy of a therapeutic agent. In some implementations, the system includes processors and memory storing instructions to process first and second multi-channel fluorescence images of cardiac fibroblasts obtained before and after treatment, generate normalized single-cell profiles, compare morphological features between timepoints, and determine therapeutic efficacy based on changes in selected morphological features.

[0031] Additional features of the system may include determining whether features move toward nonfailing values; comparing features extracted from nuclei and cytoplasm; applying thresholdbased or rule-based criteria; assessing Hoechst edge intensities and actin edge intensities; providing profiles to a machine-learning system to compute phenotypic-change metrics; determining efficacy when treated CFs shift toward non-failing phenotypes; and processing additional timepoints after treatment.

[0032] Some implementations of the present invention relate to a system for supporting treatment of a subject. In some implementations, the system includes processors and memory storing instructions to process multi-channel fluorescence images of cardiac fibroblasts, generate single-cell profiles, classify cardiac fibroblasts as failing or non-failing, and determine whether a therapeutic agent should be administered based on such classification.

[0033] Additional features of the system may include providing profiles to a machine-learning system trained with heart-failure clinical labels; evaluating whether features move toward failing values; evaluating Hoechst edge-intensity and actin edge-intensity features; selecting therapies targeted to heart failure, cardiac remodeling, or fibrosis; analyzing post-treatment CFs; adjusting treatment based on morphological changes; applying feature-based thresholds for treatment decisions; and operating on CFs derived from failing or non-failing human donors.

[0034] In some implementations, any of the methods described herein may be implemented, at least in part, by a non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing device, cause the device to perform any of the operations, steps, or functionalities described in the present disclosure. Likewise, any of the systems and subsystems described herein may be realized or supplemented by such instructions stored on a non-transitory computer-readable medium.

[0035] The foregoing description provides an overview of certain features and technical advantages of examples of the present disclosure to facilitate a better understanding of the detailed description that follows. Additional features, variations, and advantages will be described below. The concepts and specific examples disclosed herein may be utilized as a basis for modifying or designing other structures or methods to achieve the same or similar purposes as those of the present disclosure. Equivalent constructions, implementations, and variations are considered to fall within the scope of the appended claims. Characteristics of the concepts disclosed herein, including their organization, method of operation, and associated advantages, will be further understood from the following description when considered with the accompanying drawings. Each of the figures is provided for illustration and explanation only and is not intended to limit the scope of the claims.

[0036] BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The appended drawings are provided to facilitate a more detailed understanding of the features of the present disclosure and to support the following specific description. The disclosure is described below with reference to certain aspects, some of which are illustrated in the appended drawings. It should be understood that the appended drawings depict only illustrative examples of the disclosure and are therefore not to be considered as limiting its scope, as the description may encompass other equally effective implementations. In some instances, the same reference numbers in different drawings may be used to identify the same or similar elements for ease of understanding.

[0038] FIG. 1 is a block diagram in accordance with some aspects of the present invention.

[0039] FIG. 2 is a flowchart in accordance with some aspects of the present invention.

[0040] FIG. 3 is a flowchart in accordance with some aspects of the present invention.

[0041] FIG. 4 is a flowchart in accordance with some aspects of the present invention.

[0042] FIG. 5 is a block diagram in accordance with some aspects of the present invention. FIG. 6A includes representative picrosirius red (PSR) stained LV sections from non-failing and failing hearts (left); bright-field microscopy and birefringence of the PSR signal using polarized light microscopy (middle) at scale bar = 500 pm; representative images of isolated and cultured CFs stained with antibodies for a-smooth muscle actin (a -SMA, red) and fibronectin (FN, green), as well as DAPI to reveal nuclei (blue) (right) at a scale bar = 200 pm.

[0043] FIG. 6B includes images of CFs that were fixed, permeabilized and stained with Hoechst 33342 (nuclei), concanavalin A (endoplasmic reticulum [ER]), SYTO14, wheat germ agglutinin (WGA; plasma membrane and Golgi apparatus), MitoTracker (mitochondria), and Phalloidin (F-actin) depicting individual channel and overlaid composite images of the staining

[0044] FIG. 6C is a precision-recall curve fortraining and testing data evidencing high performance for predicting failing CFs.

[0045] FIG. 6D is a graphical display illustrating the level of importance of each of the input morphology features for making predictions using the ML system.

[0046] FIG. 6E is a heatmap visualization of logistic regression coefficients.

[0047] FIG. 6F is a graph of the accuracy scores for individual hearts in training and testing splits evidencing that the ML system generalizes across heart failure patients.

[0048] FIG. 6G is a collection of images of stained single cells across all organelles.

[0049] FIG. 6H is a graph of precision-recall curves for assessing generalizability of three prediction models in patient samples the model has never seen before. The three models include: 1) all features; 2) training the original model using only F-actin features; and 3) all features except F- actin features. This indicates that the machine learning model can accurately predict non-failing and failing CFs from patients it had not previously encountered. Furthermore, this also indicates that more organelles / compartments contribute to failing CF phenotypes beyond standard actin- based imaging.

[0050] FIG. 6I is a graphical display of probability estimates of failing CFs output from the ML system during experimentation.

[0051] FIG. 6J is a graphical display of a Uniform Manifold Approximation (UMAP) algorithm applied to failing single-cells treated with TGF-p receptor inhibitor or vehicle control showing a shift toward a healthier phenotype.

[0052] FIG. 7 includes graphs depicting that the “Rest” model consistently performed better than the Actin model in predicting failing or non-failing cardiac fibroblasts from individual hearts. The dotted red line indicates the average accuracy for all hearts in the “All features” model (Accuracy = 0.88). The performance metrics shown include the hearts in the training, testing, and holdout sets.

[0053] Fig. 8 is a graphical display of feature importance. DETAILED DESCRIPTION OF THE INVENTION

[0054] In the following detailed description of the present invention, reference is made to the accompanying drawings, which form a part thereof, and within which are shown by way of illustration specific embodiments by which the invention may be practiced. Numerous specific details are set forth to provide a thorough description of the embodiments of the present invention. It will be appreciated that the embodiments described herein are illustrative and not limiting. Features, functions, elements, and components described in connection with any embodiment may be combined with features, functions, elements, and components of other embodiments, in whole or in part, unless otherwise stated. Likewise, individual features may be implemented independently of other features, or in different combinations, as would be understood by a person of ordinary skill in the art. The invention therefore encompasses all variations, modifications, and equivalents that fall within the scope of the appended claims, including embodiments having any combination of the features described herein. It is to be understood that other embodiments may be utilized, and structural changes may be made without departing from the scope of the invention.

[0055] The relevant descriptions of such features may apply equally to the features and related components among all the drawings. For example, any suitable combination of the features, and variations of the same, described with components illustrated in FIG. 1 , can be employed with the components of FIG. 2, and vice versa. This pattern of disclosure applies equally to further embodiments depicted in subsequent figures and described hereinafter. It should be understood that the figures presented are not meant to be illustrative of actual views of any particular portion of the actual structure or method but are merely idealized representations employed to more clearly and fully depict the present invention defined by the claims below.

[0056] As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” include plural referents unless the content clearly dictates otherwise. As used in this specification and the appended claims, the term “or” is generally employed in its sense including “and / or” unless the context clearly dictates otherwise.

[0057] The phrases “in some embodiments,” “according to some embodiments,” “in the embodiments shown,” “in other embodiments,” and the like generally mean the particular feature, structure, or characteristic following the phrase is included in at least one implementation. In addition, such phrases do not necessarily refer to the same embodiments or different embodiments.

[0058] When language similar to “at least one of A, B, or C” or “at least one of A, B, and C” is used in the specification or claims, the phrase is intended to mean any of the following: (1) at least one of A; (2) at least one of B; (3) at least one of C; (4) at least one of A and at least one of B; (5) at least one of B and at least one of C; (6) at least one of A and at least one of C; or (7) at least one of A, at least one of B, and at least one of C. As used herein, the phrase “satisfying a threshold” may, depending on the context, refer to a value that is greater than, greater than or equal to, less than, less than or equal to, equal to, not equal to, or otherwise compared to the threshold, as appropriate.

[0059] Also, as used herein, the terms “coupled,” “coupling,” or any other variation thereof, are intended to cover a physical connection, an electrical connection, a magnetic connection, an optical connection, a communicative connection, a functional connection, a thermal connection, and / or any other connection.

[0060] Various aspects of the present disclosure are described below with reference to different apparatuses, systems, and techniques. These apparatuses, systems, and techniques are discussed in the following detailed description and are illustrated in the accompanying drawings by way of blocks, modules, components, circuits, steps, processes, algorithms, and / or similar structures (collectively referred to as “elements”). These elements may be implemented using hardware, software, firmware, or any combination thereof. Whether such elements are implemented as hardware, software, or otherwise depends on the particular application, implementation, and design constraints of the overall system.

[0061] As used herein, the term “module” should be interpreted broadly to include hardware, firmware, software, and / or any combination thereof. Likewise, as used herein, the term “processor” may refer to an implementation in hardware, firmware, software, and / or any combination thereof.

[0062] Referring now to the specifics of the present invention, some embodiments, include one or more computer systems having a memory, a user interface with a visual display (also referred to as a “graphical user interface” or “GUI”) , and a processor for executing one or more programs performing at least the steps described herein. In some embodiments, the present invention is a computer executable method or is a method embodied in software for executing the steps described herein. Further explanation of the hardware and software can be found in the Hardware and software infrastructure examples section below.

[0063] The present invention includes a system and method for distinguishing healthy cardiac fibroblasts (CFs) from failing CFs, diagnosing and treating heart failure (HF), predicting HF, and / or screening the efficacy of treating HF using a combination of advanced imaging techniques and machine learning (ML) systems and / or algorithms. As will be explained herein, the present invention enables high-throughput analysis of cellular features that characterize the health state of CFs, providing significant improvements in diagnosing and understanding cardiac fibrosis, which is a condition associated with HF.

[0064] FIG. 1 illustrates an example computing environment suitable for implementing aspects of the disclosure. The environment includes a computer system 102 in communication with an imaging device 104 that is configured to image CFs. The computer system 102 can be any known system having a processor, memory, a user interface, and the ability to communicate through known means with the imaging device 104 and / or a database 106 configured to store images of CFs. Similarly, the database 106 may be any known system configured to store and transfer data, such as the images of the CFs.

[0065] The imaging device 104 may be a high-content imaging system designed to capture detailed, multi-channel images of cellular structures with high resolution (e.g., 20x and 40x magnification) and throughput. The imaging device 104 enables precise visualization and quantification of various cellular components stained with fluorescent dyes, making it suitable for complex cellular analysis such as identifying morphological features (such as those listed in the table under the Morphological features section below) indicative of health or disease states. As a non-limiting example, the imaging device 104 may be the Celllnsight CX7 produced by Thermo Fisher Scientific. However, it should be understood that alternative imaging devices having similar functionality can be used to acquire images of the patient’s tissue and / or cells.

[0066] Some embodiments of the system further include a Stainer or “auto Stainer” 108 that is configured to stain the acquired cells in a consistent manner. However, manual staining devices and approaches may be employed.

[0067] Referring now to FIG. 2, some aspects of the present invention include a method for identifying morphological features indicative of a CF state, such as failing or non-failing. In some aspects, the method further includes distinguishing healthy CFs from failing CFs at step 212, diagnosing HF at step 214, predicting HF at step 216, screening the efficacy of treating HF and / or CFs at step 218, and / or treating the patient at step 220. As exemplified in FIG. 2, some embodiments of the method include acquiring CFs from one or more subjects at step 202; staining the CFs at step 204; performing image analysis of the CFs at step 206; providing the image analysis data to an ML system at step 208; and identifying morphological features indicative of a CF state at step 210.

[0068] At step 202, the CFs are acquired from the heart of one or more subjects. In some embodiments, the CFs are acquired from patients with known heart conditions, e.g., CFs can be acquired from subjects with HF and CFs can be acquired from subjects with non-failing hearts.

[0069] The CFs are then stained and imaged at step 204. One or more stains are applied to the CFs to highlight different cellular components. The staining enables the visualization of multiple cellular organelles and compartments in the CFs, including but not limited to the nuclei, actin cytoskeleton, endoplasmic reticulum, plasma membrane, Golgi apparatus, and mitochondrial network. By marking these structures, the present invention can visualize and extract information from the CFs through the imaging device 104 by capturing images across multiple fluorescence channels, where each channel visualizes one stain or marker. The multi-channel imaging thereby produces data suitable for quantitative morphological analysis of individual cells. In some embodiments, CFs are subjected to simultaneous staining with five fluorescent dyes to mark six cellular components or organelles such as those enumerated in the experimentation section below (see e.g., FIG. 6B). However, the number and identity of dyes may vary, and the staining protocol may employ more or fewer dyes, or may mark more or fewer cellular components or organelles, depending on the desired implementation. Any combination of stains capable of generating multi-channel images suitable for extracting morphological features is contemplated.

[0070] In some aspects, the present invention leverages a technique known multiplexed fluorescence imaging, including but not limited to approaches commonly referred to as Cell Painting, to stain and image the CFs at step 204. In some embodiments, the staining protocol departs from traditional Cell Painting dye sets to better capture morphological characteristics relevant to CF state. For example, nucleolar and cytoplasmic RNA dyes may be replaced, omitted, or supplemented with a stain configured to highlight actin fibers, thereby enhancing the detection of cytoskeletal features associated with failing CF states. In some embodiments, the actin channel includes any suitable F-actin stain, including but not limited to phalloidin-based dyes (such as Alexa Fluor 488, 568, 594, or 647 conjugates, or rhodamine- or FITC-phalloidin), livecell compatible actin probes (such as SiR-Actin or other silicon-rhodamine derivatives), or fluorescently labeled antibodies or affinity reagents that selectively bind cytoskeletal actin. Other stains or probes capable of highlighting cytoskeletal structures may also be employed.

[0071] As provided in FIG. 2, the method of the present invention further includes analyzing the images of the stained CFs at step 206. In some aspects, step 206 is executed using an imageprocessing module 300, as exemplified in FIG. 3, which is configured to process multi-channel images of the CFs. The image-processing module 300 may include functionality for adjusting, selecting, and individually processing each fluorescence channel.

[0072] In some embodiments, the image-processing module 300 may be implemented using any suitable image-analysis software, hardware, or algorithmic framework. As shown in FIG. 3, in some aspects, the image-processing module 300 may include an illumination correction module 302, a segmentation module 304, a feature extraction module 306, a filtration module 308, a curation module 310, and a normalization module 312. For example, the illumination correction module 302, segmentation module 304, and feature extraction module 306 may be implemented using software such as CellProfiler, which is configured to perform illumination correction, identify and segment single-cell nuclei and cytoplasm, and extract morphological features from each individual cell. The filtration module 308 may incorporate or interface with quality-control tools such as coSMicQC for filtering segmentation errors and debris, while the curation module 310 may utilize tools such as CytoTable for organizing and structuring singlecell feature data. The normalization module 312 may employ profiling tools such as Pycytominer to normalize and filter single-cell profiles for downstream analysis. In some aspects, one or more of these submodules (302, 304, 306, 308, 310, 312) may be realized partially or entirely through hardware, custom firmware, embedded processors, distributed computing systems, or ML-based image-analysis frameworks configured to perform the illumination correction, segmentation, feature-extraction, error-filtering, curation, or normalization steps described herein. In addition, one or more of these submodules (302, 304, 306, 308, 310, 312) may be configured to perform one or more of the sub-steps 400. Thus, the image-processing module 300 may employ any combination of software-, hardware-, or algorithm-based resources suitable for executing the sub-steps 400 using one or more of the submodules (302, 304, 306, 308, 310, 312).

[0073] As shown in the flow chart of FIG. 4, step 206 may include a series of sub-steps 400 executed by the image-processing module 300, including but not limited to correcting illumination artifacts at step 402, identifying and segmenting single-cell nuclei and cytoplasm (or other cellular regions) at step 404, and extracting a plurality of morphological features (such as those listed in the table under the Morphological Features section below) from each individual cell at step 406. In some aspects, the image-processing module 300 further executes additional operations, including filtering segmentation errors and debris at step 408, curating single-cell feature data at step 410, and normalizing and filtering single-cell profiles at step 412 to generate standardized morphological signatures.

[0074] Referring now to FIGs. 3 and 4, in some aspects, the image-processing module 300 may include an illumination correction module 302 configured to correct illumination artifacts in the multi-channel fluorescence images at step 402. Illumination artifacts may include non-uniform lighting, shading, gradients, and optical distortions that cause portions of an image to appear artificially bright or dim. Correcting these artifacts ensures accurate quantitative feature extraction by maintaining consistent illumination across the field of view. The illumination correction module 302 may employ any suitable technique, such as flat-field correction, background estimation, normalization, or ML-based illumination modeling, to correct these inconsistencies so that fluorescence intensities reflect actual biological differences rather than technical variability introduced during imaging. In some aspects, software such as CellProfiler may be used to implement illumination correction, although the module 302 may be realized in hardware, software, firmware, or any combination thereof. Variations of the illumination correction module may incorporate additional or alternative algorithms, data structures, or processing architectures to accommodate different imaging systems, image formats, or analysis requirements.

[0075] In some aspects, the image-processing module 300 may also include segmentation module 304 for segmenting single-cell nuclei and cytoplasm at step 404. Segmenting single-cell nuclei and cytoplasm can be employed for analyzing each cell individually rather than analyzing the image as a whole. Thus, the method may include identifying and isolating individual cells and their components (e.g., nuclei, cytoplasm, or other subcellular structures) through manual, automated, or hybrid approaches. In some embodiments, the image-processing module 300 utilizes one or more algorithms, such as threshold-based segmentation, edge detection, watershed algorithms, region-growing, ML segmentation, or deep-learning models, to recognize boundaries of single-cell nuclei and the surrounding cytoplasm based on differences in staining patterns, texture, and / or intensity, and to perform the segmentation step. The segmentation process may be implemented using proprietary software, commercial imaging systems, cloud-based analysis services, embedded hardware, or any other computing platform capable of performing the described functions.

[0076] In some aspects, the image-processing module 300 may include a feature extraction module 306. The feature extraction module 306 is configured to extract a plurality of morphological features from each individual cell at step 406. Morphological features, such as those listed in the table under the Morphological features section below, include quantitative measurements that describe the shape and size (e.g., area, perimeter, aspect ratio, etc.), texture (e.g., granularity and / or uniformity of staining), intensity (e.g., mean fluorescence intensity and / or signal distribution), and / or spatial organization of cellular components (e.g., nuclear positioning within the cytoplasm and / or relationships between organelles). In some embodiments, the image-processing module 300 is configured to extract over 2,000 different features for each cell, providing a detailed and multidimensional profile of the cell’s phenotype. During experimentation, more than 2,000 morphological features per cell were extracted, although the number, identity, and type of extracted features may vary depending on the implementation and may include hundreds, thousands, or any other number of features suitable fordownstream analysis.

[0077] In some aspects, the image-processing module 300 is configured to extract specific morphological features. During experimentation, it was determined that certain morphological features extracted during step 206 are unexpectedly highly predictive for distinguishing failing CFs from non-failing CFs as depicted in Fig. 8. These highly predictive features include but are not limited to: the mean intensity at the edge of Hoechst within the nucleus (referred to as “Nuclei_lntensity_MeanlntensityEdge_Hoechst” in the table of morphological features); the minimum intensity at the edge of Hoechst within the nucleus (referred to as “Nuclei_lntensity_MinlntensityEdge_Hoechst” in the table of morphological features); the total integrated intensity of actin at the edge of the cytoplasm (referred to as “Cells_lntensity_lntegratedlntensityEdge_Actin” in the table of morphological features); and the maximum intensity at the edge of Hoechst within the cytoplasm (referred to as “Cytoplasm_lntensity_MaxlntensityEdge_Hoechst” in the table of morphological features). In some aspects, other such features that correspond to edge-intensity measurements that capture subtle variations in nuclear chromatin distribution, cytoplasmic boundary organization, and actin-associated stress fiber morphology may be extracted.

[0078] The image analysis pipeline of the present invention may further include a step of filtering out segmentation errors and other debris at step 408, which may be executed by the filtration module 308. Segmentation errors occur when the software incorrectly identifies or fails to properly delineate objects (e.g., nuclei, cytoplasm) in the images. This can result in merging of adjacent cells into one object, fragmentation of a single cell into multiple parts, and / or misidentification of cellular regions, which can result in inaccurate data about cell features, compromising the reliability of the analysis. With respect to debris, during imaging, debris such as dust particles, staining artifacts, or imaging noise can appear in the images. These artifacts do not represent biological structures but can be mistakenly analyzed as cells. Including debris in the analysis can skew the results and reduce the overall quality and interpretability of the data. Thus, the present invention may employ a tool or algorithm configured to enhance the quality of data by removing segmentation errors and unwanted artifacts such as debris.

[0079] In some embodiments, the image analysis pipeline further includes curation and normalization operations performed by the curation module 310 and the normalization module 312 of the image-processing module 300. For example, the curation module 310 may utilize software such as CytoTable to organize, structure, and compile single-cell feature outputs generated during feature extraction. The normalization module 312 may employ tools such as Pycytominer to perform the image-based profiling pipeline resulting in normalized and filtered single-cell profiles. This process normalizes all single-cell profiles and filters uninformative features based on low variance and high correlations with other measurements. This process generates standardized and quality-controlled single-cell profiles suitable for downstream analysis.

[0080] The resulting data set of morphological features can be referred to as “ML input data 314” and may be in the form of tabular numerical datasets. The input data 314 include quantitative measurements from the segmented images for each CF. In some aspects, the input data 314 may be provided as inputs to the one or more ML systems 500. As noted herein, the input data 314 may include but are not limited to nucleus-related features (e.g., intensity, shape, and texture of nuclear staining); actin-related features (e.g., intensity and organization of actin fibers); and / or other organelles (e.g., features from mitochondria, endoplasmic reticulum (ER), plasma membrane, and Golgi apparatus).

[0081] As previously noted, in some aspects, the present invention includes one or more ML systems 500 as exemplified in FIG. 5. For the sake of brevity, the remainder of this disclosure will refer to a single ML system (unless specified otherwise) with the understanding that multiple ML systems can be used. The ML system 500 may reside on the computer device 102 or on another system configured to communicate with the computer system 102. The ML system 500 includes a computational framework designed to automatically learn and improve from experience without being explicitly programmed for every specific task. It may include algorithms and statistical models that process and analyze data to identify patterns, make predictions, or perform decision-making tasks. In some embodiments, the ML system 500 is trained on labeled data to map inputs to outputs. However, the ML system 500 may be configured to identify patterns or structures in unlabeled data and / or may be configured to learn by interacting with an environment and receiving feedback in the form of rewards or penalties.

[0082] In some embodiments of the present invention, as exemplified in FIG. 5, the method includes training a ML system by providing the input data to the ML system 500. To evaluate the performance of the model, the input data may be separated into training data 502 and testing data 504, with training data used during training and testing data used to evaluate the performance of the system. In some embodiments, a majority of the input data is used for training while the remainder is used for testing. For example, 70% of the input data can be training data and 30% can be testing data.

[0083] During training, the ML system 500 receives the input data and produces output data in the form of labels classifying CFs. In some aspects, the ML system 500 is trained using a supervised learning approach, where the training input data 502 is in the form of labeled data (morphological features with corresponding "failing" or "non-failing" CF labels). The ML system 500 processes the training input data 502 to generate training outputs 506, which represent predicted labels. The training outputs 506 are evaluated by an evaluation module 508 that computes loss values or performance metrics. The evaluation results are provided back to the ML system 500 to update the parameters of the ML system 500, thereby improving the model’s ability to classify cells based on their phenotypic features.

[0084] In some embodiments, the ML system 500 uses penalized logistic regression during training. This algorithm assigns weights to features to predict the probability of a cell being from a failing or non-failing heart. Penalty terms are used to prevent overfitting by discouraging overly complex models. The algorithm iteratively adjusts the feature weights to minimize the error between the predicted and actual labels in the training data. The output is a set of coefficients for each feature, indicating their contribution to the classification.

[0085] The ML system 500 is sufficiently trained to detect CF states once the ML system 500 reaches a predetermined threshold for accurately labeling the CFs in the testing data as “failing” or “nonfailing.” In some embodiments, the performance of the ML system 500 is evaluated using a precision-recall curve and an area under precision-recall curve (AUPR) score. A high AUPR score (e.g., 96% for training and 89% for testing) indicates good performance, i.e., the ML system’s ability to distinguish failing from non-failing cells. The performance threshold could be a certain score (e.g., greater than or equal to 80%, 85%, 90%, or 95%) indicative of a trained system.

[0086] In some embodiments, the ML system 500 is further evaluated on holdout input data that was not used during training. This holdout data (“ML testing data 504”) is evaluated to ensure the ML system 500 generalizes to entirely new samples that were collected in entirely separate experiments. In some aspects, one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and / or the maximum intensity at the edge of Hoechst within the cytoplasm may be emphasized, weighted, or preferentially selected by the imageprocessing module 300 or by the ML system 500 during training. In certain embodiments, the classification of CFs may be performed using a subset of features that consists essentially of the features identified above. In other embodiments, the ML system 500 automatically identifies these features through feature-ranking or feature-selection algorithms. These embodiments reflect that edge-intensity-based metrics (e.g., nuclear perimeter intensity, cytoplasmic boundary intensity, and actin edge-integrated intensity) represent core phenotypic signatures associated with failing CFs.

[0087] In some embodiments, the ML system 500 includes functionality for ranking, selecting, or weighting morphological features based on their predictive contribution. In certain aspects, the ML system 500 may assign larger coefficient values, weights, or feature-importance scores to these features as part of a penalized regression, tree-based model, or other supervised- learning architecture. For example, during training, the ML system 500 may identify that one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and / or the maximum intensity at the edge of Hoechst within the cytoplasm provide superior discriminatory power for distinguishing failing from non-failing CFs.

[0088] In other embodiments, the ML system 500 may utilize an embedded or external featureselection module configured to reduce the total feature set to a smaller subset that includes one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and / or the maximum intensity at the edge of Hoechst within the cytoplasm. In some embodiments, the ML system 500 performs classification using a reduced feature vector that comprises only edge-intensity-related morphological features, including but not limited to the four features identified above.

[0089] Although the ML system 500 provides one implementation for classifying CFs, in some embodiments the system may employ deterministic or rule-based decision processes derived from the highly predictive morphological features described above. For example, in certain aspects, a CF may be classified as failing when its minimum intensity at the edge of Hoechst within the nucleus value falls below a predetermined threshold, or when its total integrated intensity of actin at the edge of the cytoplasm value exceeds a predetermined statistical cutoff associated with abnormal actin-edge intensity. Similar threshold-based or scoring-based rules may be applied using the mean intensity at the edge of Hoechst within the nucleus or the maximum intensity at the edge of Hoechst within the cytoplasm. In some aspects, the present invention includes a ML system 500 that is trained to diagnose or predict HF status in a subject based on morphological features extracted from CFs. The ML system 500 may receive training data comprising single-cell profiles generated from CFs obtained from subjects clinically characterized as having HF or not having HF. The single-cell profiles may include any combination of the morphological features described herein, including features extracted from the nucleus, cytoplasm, and other subcellular structures.

[0090] In certain embodiments, each single-cell profile in the training data is associated with a clinical label indicating whether the CFs were obtained from cardiac tissue of a subject diagnosed with HF or from a subject without HF. These clinical labels may be derived from medical records, diagnostic assessments, imaging data, or other clinical determinations. The ML system 500 may use these labels to learn associations between the extracted morphological features and the clinical HF status of the subjects from whom the CFs were obtained.

[0091] In some aspects, the ML system 500 processes the training input data to generate predicted clinical labels for the CFs and compares the predicted labels to the clinical labels to compute one or more evaluation metrics. The evaluation metrics may include loss values, accuracy values, precision-recall metrics, or any other statistical measures. Based on the evaluation metrics, the ML system 500 may adjust internal model parameters through one or more optimization procedures. This process may be repeated for multiple training iterations until performance reaches a predetermined threshold.

[0092] In some embodiments, the morphological features used for HF prediction / diagnosis include features that were observed to be highly predictive of CF phenotype, such as the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm. These features, individually or in combination, may provide signals that correspond to remodeling, stress responses, or other cellular characteristics associated with HF.

[0093] In further embodiments, the trained ML system 500 may be used to generate diagnoses or predictions of HF risk, likelihood, or severity for new subjects by receiving single-cell profiles extracted from CFs obtained from the subjects. The ML system 500 may output a diagnosis or predicted label, probability value, or score indicative of whether the subject exhibits a cellular phenotype consistent with HF.

[0094] In additional embodiments, the system may incorporate both morphological feature data and clinical metadata during training. For example, training data may include single-cell profiles in combination with subject age, sex, comorbidities, medication history, or other clinical variables that influence HF status. The ML system 500 may use such combined data to improve predictive performance and robustness. Some embodiments further include a treatment step of treating the patient with a therapeutic if the diagnosis indicates that treatment is necessary. The patient may be treated with therapeutics that improve CFs or HF, including but not limited to inhibitors of the TGF-p pathway alone or in combination with other therapeutics. Non-limiting examples of TGF-p receptor inhibitors include SB525334, pirfenidone, Galunisertib (LY2157299), Vactosertib (TEW-7197), LY3200882 and NIS793.

[0095] SB525334 is a potent and selective inhibitor of the TGF-p type I receptor (ALK5) and has been utilized extensively in preclinical studies to investigate the role of TGF-p signaling in various diseases including cancer and fibrosis. The mechanism of action of SB525334 involves inhibition of TGF-p receptor kinase activity, which prevents phosphorylation of downstream signaling molecules, specifically the Smad proteins. Upon TGF-p binding to its receptor, Smad2 and Smad3 are phosphorylated, leading to their translocation to the nucleus where they regulate the expression of fibrotic genes. By inhibiting this receptor, SB525334 effectively disrupts the TGF-p / Smad signaling cascade, thereby mitigating myofibroblast activation and the fibrotic response. The investigations into this inhibitor as a tool compound in connection with the present invention has revealed that inhibition of the TGF-p / Smad signaling cascade by treatment of failing human cardiac fibroblasts with SB525334 effectively reverts these cells to a quiescent state reminiscent of non-failing fibroblasts, suggesting its potential efficacy in ameliorating cardiac fibrosis in human heart failure patients.

[0096] Another promising agent is pirfenidone, which has been shown in preclinical studies to alleviate cardiac fibrosis induced by pressure overload through inhibition of the TGF-pi / Smad3 signaling pathway. Pifenidone is already approved for the treatment of idiopathic pulmonary fibrosis and its mechanism involves the modulation of profibrotic cytokines and the reduction of collagen synthesis, making it a valuable candidate for treating cardiac fibrosis.

[0097] Galunisertib (LY2157299) is a small-molecule inhibitor of the TGF-p type I receptor kinase and prevents the phosphorylation of SMAD2 and SMAD3, critical mediators of TGF-p signaling. Galunisertib has demonstrated promising activity in phase II trials for patients with advanced hepatocellular carcinoma and has been evaluated in combination with other therapies to enhance immune responses or sensitize tumors to chemotherapy. Preclinical studies have shown that Galunisertib reduces fibrosis in models of pulmonary fibrosis, although it has not yet reached clinical application.

[0098] Vactosertib (TEW-7197), a small molecule inhibitor that specifically targets the TGF-p type I receptor, has emerged as a promising anti-fibrotic agent. Preclinical studies have demonstrated that Vactosertib can significantly reduce fibrosis in various models, including those of pulmonary and cardiac fibrosis. Vactosertib is currently in clinical trials for a variety of conditions where TGF-p signaling plays a role, including solid tumors, to overcome immune checkpoint blockade and in hepatocellular carcinoma. The safety profile of Vactosertib has also been the focus of recent research, with studies indicating that it is well tolerated in clinical settings. Its oral bioavailability and manageable side effects also make it a suitable candidate for long-term treatment regimen options.

[0099] LY3200882 is a novel, highly selective small molecule inhibitor that targets the TGF-p type I receptor and was developed to provide more potent and selective inhibition compared with earlier inhibitors such as Galunisertib. LY3200882 functions as an ATP-competitive inhibitor of the serine-threonine kinase domain of this receptor. Preclinical studies have demonstrated that LY3200882 exhibits a strong inhibitory effect on the TGF-p receptor and reduces collagen deposition and improves tissue function. Clinical trials have begun to evaluate the safety and efficacy of LY3200882 in patients with advanced cancer. Finally, the ability of LY3200882 to modulate the TGF-p signaling pathway may also have implications for improving outcomes in patients with fibrotic diseases by reducing the activation of fibroblasts into myofibroblasts.

[0100] NIS793 is a human monoclonal antibody that specifically neutralizes TGF-p and prevents it from interacting with its receptor on target cells, effectively disrupting the downstream signaling pathways mediated by Smad proteins. Clinical trials have begun to evaluate the safety and efficacy of NIS793 in patients with advanced tumors, where it is being tested in combination with other therapies such as immune checkpoint inhibitors. In preclinical studies, NIS793 has shown promising results in reducing fibrosis and improving lung function in animal models of pulmonary fibrosis and has been shown to attenuate the fibrotic response in liver fibrosis models.

[0101] The method of diagnosing, which is captured in FIG. 2, includes acquiring CFs from a patient at step 202, staining the CFs at step 204, performing the one or more steps in the image analysis pipeline described above at step 206, and then providing the input data corresponding to the patient’s CFs to the trained ML system at step 208. The trained ML system then outputs to a medical provider and / or patient whether or not the CFs are in a “failing” or “non-failing” state at step 212. The outputs from the ML system could also be in the form of a percent likelihood of failure or any other form to help diagnose HF at step 216. As noted above, the method may further include a treatment step 220 of treating the patient with a therapeutic (such as those disclosed above) if the diagnosis indicates that treatment is necessary.

[0102] In some embodiments, the method of the present invention includes providing input data to a trained ML system to identify the effectiveness of therapeutics at step 218. Again, the ML system may be trained in accordance with the method described herein. As a drug screening tool, the ML system can screen for potential treatment protocols, such as anti-fibrotic drugs, by observing changes in cell morphology after treatment.

[0103] The method of screening includes acquiring CFs from a patient undergoing treatment, staining the CFs, performing the one or more steps in the image analysis pipeline described above, and then providing the input data corresponding to the patient’s CFs to the trained ML system. The trained ML system then outputs to a medical provider and / or patient whether or not the CFs are in a “failing” or “non-failing” state or any other form to help determine the state of HF. In some embodiments, the method is performed at least once prior to treatment and at least once during or following treatment to determine whether there are any changes in the CFs. The efficacy of the treatment can then be provided to the medical provider and / or patient.

[0104] Some embodiments further include a dimensionality reduction technique, such as Uniform Manifold Approximation and Projection (UMAP) to visualize the high-dimensional feature space. This allows for a clear separation between failing and non-failing CFs, with treated failing CFs clustering closer to non-failing cells. The UMAP visualization provides a valuable tool for interpreting cellular changes across different treatments and health states.

[0105] In some embodiments, the present invention includes methods for evaluating the phenotypic effects of a therapeutic agent, a candidate compound, a biologic, or any substance applied to cardiac fibroblasts. The system may determine whether exposure to the agent induces measurable changes in one or more morphological features extracted from the single-cell profiles generated as described herein. Such embodiments enable the system to detect subtle or early phenotypic responses of cardiac fibroblasts to chemical perturbation.

[0106] In certain embodiments, the system is configured to screen one or more candidate compounds by comparing morphological features extracted from cardiac fibroblasts exposed to each candidate compound with corresponding features extracted from untreated or baseline cardiac fibroblasts. The system may identify a candidate compound as active, inactive, or partially active based on whether the compound induces changes in one or more morphological features indicative of phenotypic improvement, phenotypic worsening, or other measurable cellular responses.

[0107] In some aspects, the system evaluates changes in highly predictive morphological features, including but not limited to: (i) the mean intensity at the edge of Hoechst within the nucleus; (ii) the minimum intensity at the edge of Hoechst within the nucleus; (iii) the total integrated intensity of actin at the edge of the cytoplasm; and (iv) the maximum intensity at the edge of Hoechst within the cytoplasm. Changes in one or more of these features may be used to determine whether a candidate compound shifts cardiac fibroblast morphology toward or away from phenotypes characteristic of non-failing or failing cardiac states.

[0108] In some embodiments, the first and second sets of normalized single-cell profiles generated from untreated and treated cardiac fibroblasts are provided to the ML system 500 which is further configured to compute a score, probability, similarity value, cluster assignment, reduced- dimensionality representation, or other quantitative metric indicative of the phenotypic effect of the candidate compound. The system may identify a compound as active when the metric indicates that the morphology of treated cardiac fibroblasts has shifted toward a phenotype associated with non-failing cardiac fibroblasts or away from a phenotype associated with failing cardiac fibroblasts.

[0109] In additional embodiments, the determination of compound activity or efficacy is performed without machine learning, and instead uses one or more threshold-based, rule-based, or algorithmic criteria. For example, a compound may be identified as effective when the minimum intensity at the edge of Hoechst within the nucleus exceeds a predetermined threshold, orwhen the total integrated intensity of actin at the edge of the cytoplasm decreases below a specified cutoff. These deterministic approaches may be used alone or in combination with ML-assisted approaches.

[0110] In some embodiments, the system screens candidate compounds at a plurality of doses, time points, or experimental conditions, and evaluates how the morphological features change across these conditions. In certain aspects, the system generates a dose-response curve, concentration-response relationship, or temporal trajectory based on the morphological changes induced by the compound.

[0111] In additional embodiments, the system ranks or prioritizes multiple candidate compounds based on their phenotypic effects on cardiac fibroblasts. Rankings may be based on the magnitude, direction, or pattern of morphological feature changes, ML-derived scores or probabilities, or the degree to which a compound shifts cardiac fibroblasts toward phenotypes associated with non-failing cardiac states. Such ranked results may be used to identify lead compounds, select promising therapeutic candidates, or guide subsequent optimization.

[0112] In some aspects, the present invention also includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the nuclear intensity of stained CFs before and after treatment. In some embodiments, the method for determining the efficacy includes determining the whether the intensity of nuclei is greater before treatment and / or whether the intensity has decreased after treatment. If intensity after treatment decreases, then the treatment is deemed effective. The system and method can then notify the medical provider and / or patient.

[0113] In some aspects, the present invention also includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the actin intensity of stained CFs before and after treatment. In some embodiments, the method for determining the efficacy includes determining the whether the intensity of F-actin is greater before treatment and whether the intensity has decreased after treatment. If intensity after treatment decreases, then the treatment is deemed effective. Again, system and method can then notify the medical provider and / or patient.

[0114] In some embodiments, the method includes analyzing both the actin and the nuclear intensity of stained CFs before and after treatment. If the intensity has decreased after treatment, then the treatment is deemed effective and notify the medical provider and / or patient can be notified. In some aspects, the present invention includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the mean intensity at the edge of Hoechst within the nucleus of stained CFs before and after treatment. In some embodiments, the method includes determining whether this nuclear Hoechst edge intensity is elevated prior to treatment and whether it decreases following administration of the therapeutic agent. If the mean intensity at the nuclear edge decreases after treatment, the therapy is deemed to be effective. The system and method can then provide an indication of treatment effectiveness to the medical provider and / or patient.

[0115] In some aspects, the present invention also includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the minimum intensity at the edge of Hoechst within the nucleus of stained CFs before and after treatment. In some embodiments, the method includes determining whether this minimum nuclear edge intensity is abnormally low prior to treatment and whether it increases following administration of the therapeutic agent. If the minimum intensity at the nuclear edge increases after treatment, the therapy is deemed to be effective. The system and method can then notify the medical provider and / or patient.

[0116] In some aspects, the present invention also includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the total integrated intensity of actin at the edge of the cytoplasm before and after treatment. In some embodiments, the method includes determining whether cytoplasmic actin edge-integrated intensity is elevated prior to treatment and whether it decreases following administration of the therapeutic agent. If the total integrated actin intensity at the cytoplasmic edge decreases after treatment, then the treatment is deemed effective. The system and method can then notify the medical provider and / or patient.

[0117] In some aspects, the present invention also includes a method for determining the efficacy of a drug or therapeutic agent based on analyzing the maximum intensity at the edge of Hoechst within the cytoplasm of stained CFs before and after treatment. In some embodiments, the method includes determining whether this cytoplasmic Hoechst edge intensity is elevated prior to treatment and whether it decreases following administration of the therapeutic agent. If the maximum cytoplasmic Hoechst edge intensity decreases after treatment, then the treatment is deemed effective. The system and method can then notify the medical provider and / or patient.

[0118] Experimentation

[0119] The present invention represents a significant advance in cardiac fibrosis research, offering precision, efficiency, and adaptability for clinical and experimental applications. The invention was evaluated on 37,725 single CFs obtained from non-failing and failing human hearts. Human CFs were extracted from explanted hearts of patients with idiopathic dilated cardiomyopathy undergoing cardiac transplantation, or non-failing donor controls. Picrosirius red staining of left ventricular sections prior to CF extraction revealed profound fibrosis in heart failure (HF) patients compared to non-failing donors (see FIG. 6A). Consistent with prior findings23, CFs isolated from HF patients appeared to maintain an activated state in culture, as revealed by a- SMA staining (FIG. 6A). CFs were then subjected to Cell Painting through simultaneous staining with five fluorescent dyes to mark six cellular components or organelles (FIG. 6B)4. The Cell Painting assay was modified by dropping traditional nucleoli and cytoplasmic RNA staining in favor of a dedicated F-actin channel to more closely mirror a-SMA.

[0120] An image analysis pipeline was applied to adjust, select, and process each Cell Painting channel. Cel I Profi le r was used to adjust for illumination artifacts, segment single-cell nuclei and cytoplasm, and extract 2,022 morphological features for every single cell. Subsequently, coSMicQC was employed to filter segmentation errors and other debris, CytoTable to curate single-cell features, and Pycytominer to perform the image-based profiling pipeline. This process normalized all single-cell profiles and filtered uninformative features based on low variance and high correlations with other measurements. A machine learning pipeline was then applied, splitting 30% of the data to the testing set and 70% to training. An entire failing heart was also withheld, which represents the expectation of model performance with unseen data.

[0121] A penalized logistic regression classifier predicted failing CFs with 96% AUPR curve in the training set and 89% in the test set, which is substantially higher than a negative control baseline (FIG. 6C). The model used different input morphology features from multiple categories to make accurate predictions (FIG. 6D). The most important features from this signature related to nucleus and actin intensities, which indicates the main changes came from these structures in failing CFs (FIG. 6E). Overall, accuracies above 80% for predicting failing CFs in all hearts, including failing heart #4, which was excluded from training, was observed (FIG. 6F). Representative failing and non-failing CFs for the most important feature per organelle show substantial heterogeneity, which is accounted for by the model (FIG. 6G). Previously unencountered CFs derived from distinct failing and non-failing patients were then introduced to the pipeline and, remarkably, equivalent prediction performance was observed when applied to these new patient samples (95% AUPR), providing further evidence of robust generalizability of the model (FIG. 6H). Treatment of failing CFs with SB525334, an inhibitor of the profibrotic TGF-p receptor, resulted in a shift toward a non-failing classification, further illustrating that the model is capturing an activated CF state (FIG. 61). Lastly, applying Uniform Manifold Approximation (UMAP) to all single-cells demonstrated heterogeneous populations but also clear differences in failing and non-failing CFs, with SB525334 reverting 83% of failing cells to a non-failing state (FIG. 6J).

[0122] During experimentation, the present invention was compared to the standard practice of classifying fibroblasts as activated or non-activated based on the presence of aSMA stress fibers. To do so, a new computational analysis was performed to quantitatively compare the approach of the present invention to the standard aSMA approach. Specifically, two new machine learning models were trained: 1) using only F-actin features and 2) using all other features except F-actin features. Using F-actin features only, the following AUPR (area under the precision recall curve) performance metrics were observed: (Training Data: 0.63; Testing: 0.61). For the all-feature- except F-actin model the following AUPRs were observed: (Training: 0.90; Testing: 0.86). For reference, the original model performance was: (Training: 0.96; Testing: 0.89). This information is provided in the table below for convenience:

[0123] As applied to individual hearts, consistently improved performance for the “Everything except F-actin model (Rest)” was seen as depicted in FIG. 7.

[0124] These pre-trained models were then applied to predict failing and non-failing CFs in the new plate of data (FIG. 6H). The results were similar to the initial training / testing performance: the “All features” model performed the best (all features including actin; blue), followed by the “Rest” model (green), and lastly, the “F-actin only” model (orange).

[0125] Next, to assess how cell structures contribute to predictions beyond the traditional actin-based staining, the model was retrained in two new ways: 1) only F-actin features and 2) using all features except F-actin. Applying each of these models to the new hearts showed increased performance for the model without F-actin (AUPR=0.88) compared to the model with only F- actin (AUPR=0.43), which provides further evidence of robust generalizability and that CFs are changing dynamically in ways beyond F-actin (FIG. 6H).

[0126] Targeted therapies that block or reverse CF activation do not exist5. Phenotypic screening for agents with the ability to normalize thousands of morphological differences between failing and non-failing CFs, as opposed to using only a-SMA as a readout, should facilitate discovery of antifibrotic modalities with a higher likelihood of being successfully translated to the clinic. The Cell Painting and machine learning assay described here also has the ability to uncover novel mechanisms of CF activation. For example, nuclear intensity is the feature that most highly differentiates failing and non-failing CFs, and studies to elucidate the molecular underpinnings of this difference could reveal new therapeutic targets. Finally, the protocol described here could be employed as a diagnostic tool to stratify patients with different forms of HF, thus enabling precision therapy. This could be accomplished through invasive means, by using Cell Painting to characterize purified CFs obtained from patient myocardial biopsies, or noninvasively, by assessing the morphological fingerprints of cells exposed to serum from patients with cardiac disease due to distinct etiologies.

[0127] Morphological features

[0128] Hardware, Software, and Medium Implementations

[0129] The techniques described herein may be implemented in whole or in part using hardware, software, firmware, or any suitable combination thereof. For example, embodiments may be realized using special-purpose circuitry (e.g., application-specific integrated circuits or field- programmable gate arrays), programmable processors executing instructions, or combinations of these. Instructions stored on a non-transitory computer-readable medium, when executed by one or more processors, may cause a computing device to perform any of the steps or methods described herein.

[0130] Computing Devices and Environments

[0131] The subject matter described herein may be embodied in or executed by any suitable computing device or platform, including without limitation desktop computers, laptops, servers, tablets, smartphones, embedded controllers, or distributed / cloud-based systems. A computing device may include one or more processors, memory components, storage components, input / output interfaces, communication modules (e.g., wired or wireless transceivers), and / or display devices. Functionality may be centralized on a single device or distributed across multiple devices connected by a communication network. Input / output interfaces may include touch input, buttons, gesture-based controls, sensors, or other mechanisms as appropriate.

[0132] Computer-Readable Media and Program Code

[0133] A non-transitory computer-readable medium may include electronic, magnetic, optical, electromagnetic, or semiconductor storage devices, or any suitable combination thereof. Nonlimiting examples include hard drives, solid-state drives, RAM, ROM, flash memory, CD-ROMs, and optical or magnetic storage devices. Program code stored on such a medium may be transmitted using wired or wireless connections (including but not limited to optical fiber, RF, or other communication channels) and may be written in any suitable programming language, such as procedural or object-oriented languages.

[0134] Implementation Notes and Diagrams

[0135] Flowcharts, block diagrams, and system descriptions herein represent logical groupings of functionality rather than specific hardware arrangements. Steps described in the methods may be performed in any order, including concurrently or in parallel, and additional steps may be included or omitted as appropriate. Features described with respect to one embodiment may be combined with features of another embodiment unless otherwise indicated.

[0136] The following provides an overview of some aspects of the present disclosure:

[0137] Aspect 1 : A computer-implemented method for analyzing cardiac fibroblasts, comprising: acquiring one or more multi-channel fluorescence images of stained cardiac fibroblasts; processing, by an image-processing module, the one or more multi-channel fluorescence images, the processing comprising performing illumination correction on each fluorescence channel, segmenting nuclei, cytoplasm, and whole-cell regions within the images, extracting a plurality of morphological features from each cardiac fibroblast, filtering segmentation errors or debris, curating single-cell feature data, and normalizing the curated feature data to generate single-cell profiles; providing the single-cell profiles as input data to a machine-learning system configured to classify cardiac fibroblasts; and classifying, by the machine-learning system, one or more cardiac fibroblasts as failing or non-failing based on the input data.

[0138] Aspect 2: The method of Aspect 1 , wherein the stained cardiac fibroblasts have been stained with an F-actin stain.

[0139] Aspect 3: The method of Aspect 1 , wherein extracting the plurality of morphological features comprises extracting features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

[0140] Aspect 4: The method of Aspect 1 , wherein extracting the plurality of morphological features comprises extracting features of at least nuclei and cytoplasm.

[0141] Aspect 5: The method of Aspect 1 , wherein extracting the plurality of morphological features comprises extracting at least one of the following features: the mean intensity at the edge of Hoechst within the nucleus; the minimum intensity at the edge of Hoechst within the nucleus; the total integrated intensity of actin at the edge of the cytoplasm; and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0142] Aspect 6: The method of Aspect 1 , wherein curating the single-cell feature data comprises organizing the data into a tabular format.

[0143] Aspect 7: The method of Aspect 1 , wherein the machine-learning system has been trained via a plurality of steps, including: receiving, by the machine-learning system, training input data comprising normalized single-cell profiles generated from the extracted morphological features of a plurality of cardiac fibroblasts; receiving, by the machine-learning system, corresponding ground-truth labels indicating whether each cardiac fibroblast in the training input data was obtained from failing or non-failing cardiac tissue; processing, by the machine-learning system, the training input data to generate predicted labels for the plurality of cardiac fibroblasts; comparing the predicted labels to the ground-truth labels to compute one or more evaluation metrics indicative of training performance; and adjusting, by the machine-learning system, one or more model parameters based on the evaluation metrics such that the predictive accuracy of the machine-learning system improves over successive training iterations.

[0144] Aspect 8: The method of Aspect 1 , further comprising diagnosing heart failure in a subject based on the classification of the one or more cardiac fibroblasts as failing or non-failing. Aspect 9: The method of Aspect 1 , further comprising predicting a likelihood of heart failure in a subject based on the classification of the one or more cardiac fibroblasts as failing or nonfailing.

[0145] Aspect 10: The method of Aspect 1 , further comprising: acquiring first fluorescence images of cardiac fibroblasts obtained from a subject before administration of a therapeutic agent; acquiring second fluorescence images of cardiac fibroblasts obtained from the subject after administration of the therapeutic agent; processing, by the image-processing module, the first and second multi-channel fluorescence images to generate corresponding sets of normalized single-cell profiles, the processing comprising illumination correction, segmentation of nuclei and cytoplasm, extraction of morphological features, filtration of segmentation errors, curation of single-cell feature data, and normalization of the curated data; comparing, by a computing device, morphological features extracted from the first images to corresponding morphological features extracted from the second images; and determining the efficacy of the therapeutic agent based on changes in one or more of the morphological features between the first and second time points.

[0146] Aspect 1 1 : The method of Aspect 1 , further comprising administering a therapeutic agent to a subject if the classification indicates that the subject has failing cardiac fibroblasts.

[0147] Aspect 12: A computer-implemented method for predicting heart failure in a subject, comprising: processing, by an image-processing module, one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features extracted from the images; providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having heart failure or not having heart failure; and predicting, by the machine-learning system, whether the subject has heart failure based on the morphological features contained in the single-cell profiles.

[0148] Aspect 13: The method of Aspect 12, wherein training the machine-learning system comprises receiving training input data comprising single-cell profiles of cardiac fibroblasts associated with clinical labels indicating heart failure status.

[0149] Aspect 14: The method of Aspect 13, wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

[0150] Aspect 15: The method of Aspect 12, wherein the morphological features used for prediction comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm. Aspect 16: The method of Aspect 12, wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their predictive importance for determining heart failure.

[0151] Aspect 17: The method of Aspect 12, wherein predicting heart failure comprises generating a probability or numerical score that the subject has or will have heart failure.

[0152] Aspect 18: The method of Aspect 12, wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

[0153] Aspect 19: The method of Aspect 12, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

[0154] Aspect 20: The method of Aspect 12, wherein the extracted morphological features comprise at least nuclei and cytoplasm.

[0155] Aspect 21 : A computer-implemented method for diagnosing heart failure in a subject, comprising: processing, by an image-processing module, one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features extracted from the images; providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having heart failure or not having heart failure; and diagnosing, by the machine-learning system, whether the subject has heart failure based on the morphological features contained in the single-cell profiles.

[0156] Aspect 22: The method of Aspect 21 , wherein training the machine-learning system comprises receiving training input data comprising single-cell profiles of cardiac fibroblasts associated with clinical labels indicating heart failure status.

[0157] Aspect 23: The method of Aspect 21 , wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their diagnostic importance for determining heart failure.

[0158] Aspect 24: The method of Aspect 21 , wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

[0159] Aspect 25: The method of Aspect 21 , wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

[0160] Aspect 26: The method of Aspect 21 , wherein the morphological features used for diagnosing heart failure comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0161] Aspect 27: The method of Aspect 21 , wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

[0162] Aspect 28: A computer-implemented method for screening the efficacy of a therapeutic agent, comprising: acquiring first multi-channel fluorescence images of cardiac fibroblasts obtained from a subject or sample prior to exposure to the therapeutic agent, and acquiring second multichannel fluorescence images of cardiac fibroblasts obtained from the subject or sample following exposure to the therapeutic agent; processing, by an image-processing module, the first and second multi-channel fluorescence images to generate first and second sets of normalized single-cell profiles, the processing comprising performing illumination correction, segmenting nuclei and cytoplasm, extracting morphological features, filtering segmentation errors or debris, curating single-cell feature data, and normalizing the curated data; comparing, by a computing device, one or more morphological features from the first set of normalized single-cell profiles to corresponding morphological features from the second set of normalized single-cell profiles; and determining the efficacy of the therapeutic agent based on changes in the one or more morphological features between the first and second sets of normalized singlecell profiles.

[0163] Aspect 29: The method of Aspect 28, wherein determining the efficacy of the therapeutic agent comprises determining whether one or more morphological features move toward values characteristic of non-failing cardiac fibroblasts.

[0164] Aspect 30: The method of Aspect 28, wherein the morphological features compared comprise features extracted from at least nuclei and cytoplasm.

[0165] Aspect 31 : The method of Aspect 28, wherein determining the efficacy of the therapeutic agent comprises determining whether a morphological feature increases, decreases, falls within a predetermined range, or exceeds a predetermined threshold.

[0166] Aspect 32: The method of Aspect 28, wherein comparing the morphological features comprises comparing one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0167] Aspect 33: The method of Aspect 28, further comprising providing the first and second sets of normalized single-cell profiles to a machine-learning system configured to compute a metric indicative of phenotypic change induced by the therapeutic agent.

[0168] Aspect 34: The method of Aspect 28, wherein the therapeutic agent is determined to be effective when the machine-learning system computes a metric indicating that the second set of normalized single-cell profiles is more similar to profiles of non-failing cardiac fibroblasts than to profiles of failing cardiac fibroblasts.

[0169] Aspect 35: The method of Aspect 28, wherein determining the efficacy of the therapeutic agent comprises applying one or more rule-based, threshold-based, or algorithmic criteria to the morphological feature changes.

[0170] Aspect 36: The method of Aspect 28, further comprising acquiring additional fluorescence images of cardiac fibroblasts at one or more intermediate or later time points following exposure to the therapeutic agent.

[0171] Aspect 37: A method of treating a subject, comprising: acquiring one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject; processing, by an imageprocessing module, the images to generate single-cell profiles comprising morphological features extracted from the cardiac fibroblasts; classifying, by a machine-learning system or by one or more rule-based criteria, the cardiac fibroblasts as failing or non-failing based on the morphological features; and administering a therapeutic agent to the subject when the cardiac fibroblasts are classified as failing.

[0172] Aspect 38: The method of Aspect 37, wherein classifying the cardiac fibroblasts comprises providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles associated with clinical labels indicating heart failure status.

[0173] Aspect 39: The method of Aspect 37, wherein classifying the cardiac fibroblasts comprises determining whether one or more morphological features move toward values characteristic of failing cardiac fibroblasts.

[0174] Aspect 40: The method of Aspect 37, wherein classifying the cardiac fibroblasts comprises evaluating one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0175] Aspect 41 : The method of Aspect 37, wherein the therapeutic agent comprises a treatment indicated for heart failure, fibrosis, or cardiac remodeling.

[0176] Aspect 42: The method of Aspect 37, further comprising acquiring additional cardiac fibroblasts from the subject after administration of the therapeutic agent and evaluating changes in the morphological features to determine whether the treatment is effective.

[0177] Aspect 43: The method of Aspect 42, further comprising modifying the therapeutic agent or its dosage based on changes in the morphological features following treatment.

[0178] Aspect 44: The method of Aspect 37, wherein the therapeutic agent is administered when at least one morphological feature exceeds or falls below a predetermined cutoff. Aspect 45: The method of Aspect 37, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

[0179] Aspect 46: The method of Aspect 37, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

[0180] Aspect 47: A system for analyzing cardiac fibroblasts, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: acquire one or more multi-channel fluorescence images of stained cardiac fibroblasts; perform illumination correction on each fluorescence channel; segment nuclei, cytoplasm, and whole-cell regions within the images; extract a plurality of morphological features from each cardiac fibroblast; filter segmentation errors or debris; curate single-cell feature data; normalize the curated feature data to generate single-cell profiles; and provide the single-cell profiles to a machine-learning system configured to classify one or more cardiac fibroblasts as failing or non-failing based on the single-cell profiles.

[0181] Aspect 48: The system of Aspect 47, wherein the stained cardiac fibroblasts have been stained with an F-actin stain.

[0182] Aspect 49: The system of Aspect 47, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

[0183] Aspect 50: The system of Aspect 47, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

[0184] Aspect 51 : The system of Aspect 47, wherein the extracted morphological features comprise at least one of: the mean intensity at the edge of Hoechst within the nucleus; the minimum intensity at the edge of Hoechst within the nucleus; the total integrated intensity of actin at the edge of the cytoplasm; and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0185] Aspect 52: The system of Aspect 47, wherein curating the single-cell feature data comprises organizing the data into a tabular format.

[0186] Aspect 53: The system of Aspect 47, wherein the machine-learning system has been trained via operations comprising: receiving training input data comprising normalized single-cell profiles generated from morphological features extracted from a plurality of cardiac fibroblasts; receiving corresponding ground-truth labels indicating whether each cardiac fibroblast was obtained from failing or non-failing cardiac tissue; processing the training input data to generate predicted labels; comparing the predicted labels to the ground-truth labels to compute one or more evaluation metrics; and adjusting model parameters based on the evaluation metrics.

[0187] Aspect 54: The system of Aspect 47, further configured to diagnose heart failure in a subject based on the classification of the cardiac fibroblasts as failing or non-failing. Aspect 55: The system of Aspect 47, further configured to predict a likelihood of heart failure in a subject based on the classification of the cardiac fibroblasts as failing or non-failing.

[0188] Aspect 56: The system of Aspect 47, further configured to: acquire first fluorescence images of cardiac fibroblasts obtained from a subject before administration of a therapeutic agent and second fluorescence images after administration of the therapeutic agent; process the first and second fluorescence images to generate corresponding sets of normalized single-cell profiles; compare morphological features extracted from the first images to corresponding morphological features extracted from the second images; and determine an efficacy of the therapeutic agent based on changes in one or more morphological features.

[0189] Aspect 57: The system of Aspect 47, further configured to recommend administration of a therapeutic agent to a subject when the classification indicates that the subject has failing cardiac fibroblasts.

[0190] Aspect 58: A system for diagnosing heart failure in a subject, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features; and provide the single-cell profiles to a machine-learning system trained using single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having or not having heart failure, the machine-learning system being configured to diagnose whether the subject has heart failure based on the morphological features.

[0191] Aspect 59: The system of Aspect 58, wherein the machine-learning system is trained using training input data comprising single-cell profiles associated with clinical labels indicating heart failure status.

[0192] Aspect 60: The system of Aspect 58, wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their diagnostic importance for determining heart failure.

[0193] Aspect 61 : The system of Aspect 58, wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

[0194] Aspect 62: The system of Aspect 58, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

[0195] Aspect 63: The system of Aspect 58, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus. Aspect 64: The system of Aspect 58, wherein the morphological features used for diagnosing heart failure comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0196] Aspect 65: The system of Aspect 58, wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

[0197] Aspect 66: A system for screening the efficacy of a therapeutic agent, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process first and second multi-channel fluorescence images of cardiac fibroblasts obtained before and after exposure to a therapeutic agent to generate first and second sets of normalized single-cell profiles; compare one or more morphological features from the first set of normalized single-cell profiles to corresponding morphological features from the second set of normalized single-cell profiles; and determine the efficacy of the therapeutic agent based on changes in the one or more morphological features between the first and second sets of normalized single-cell profiles.

[0198] Aspect 67: The system of Aspect 66, wherein determining the efficacy of the therapeutic agent comprises determining whether one or more morphological features move toward values characteristic of non-failing cardiac fibroblasts.

[0199] Aspect 68: The system of Aspect 66, wherein the morphological features compared comprise features extracted from at least nuclei and cytoplasm.

[0200] Aspect 69: The system of Aspect 66, wherein determining the efficacy of the therapeutic agent comprises determining whether a morphological feature increases, decreases, falls within a predetermined range, or exceeds a predetermined threshold.

[0201] Aspect 70: The system of Aspect 66, wherein comparing the morphological features comprises comparing one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0202] Aspect 71 : The system of Aspect 66, further configured to provide the first and second sets of normalized single-cell profiles to a machine-learning system configured to compute a metric indicative of phenotypic change induced by the therapeutic agent.

[0203] Aspect 72: The system of Aspect 71 , wherein the metric comprises a score, probability, similarity value, or reduced-dimensionality representation reflecting morphological differences between the first and second sets of normalized single-cell profiles. Aspect 73: The system of Aspect 71 , wherein the therapeutic agent is determined to be effective when the machine-learning system computes a metric indicating that the second set of normalized single-cell profiles is more similar to profiles of non-failing cardiac fibroblasts than to profiles of failing cardiac fibroblasts.

[0204] Aspect 74: The system of Aspect 66, wherein determining the efficacy of the therapeutic agent comprises applying one or more rule-based, threshold-based, or algorithmic criteria to the morphological feature changes.

[0205] Aspect 75: The system of Aspect 66, further configured to process fluorescence images acquired at one or more intermediate or later time points following exposure to the therapeutic agent.

[0206] Aspect 76: A system for supporting treatment of a subject, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features; classify the cardiac fibroblasts as failing or non-failing based on the morphological features; and determine, via a treatment-decision module, whether a therapeutic agent should be administered to the subject based on the classification.

[0207] Aspect 77: The system of Aspect 76, wherein classifying the cardiac fibroblasts comprises providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles associated with clinical labels indicating heart failure status.

[0208] Aspect 78: The system of Aspect 76, wherein classifying the cardiac fibroblasts comprises determining whether one or more morphological features move toward values characteristic of failing cardiac fibroblasts.

[0209] Aspect 79: The system of Aspect 76, wherein classifying the cardiac fibroblasts comprises evaluating features extracted from at least nuclei and cytoplasm.

[0210] Aspect 80: The system of Aspect 76, wherein classifying the cardiac fibroblasts comprises evaluating one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

[0211] Aspect 81 : The system of Aspect 76, wherein the therapeutic agent recommended for administration comprises a treatment indicated for heart failure, fibrosis, or cardiac remodeling.

[0212] Aspect 82: The system of Aspect 76, wherein the cardiac fibroblasts comprise human cardiac fibroblasts derived from failing and non-failing human hearts.

[0213] Aspect 83: The system of Aspect 76, wherein the multi-channel fluorescence images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain. The foregoing disclosure provides illustration and description but is not intended to be exhaustive or to limit the Aspects to the precise forms disclosed. Modifications and variations may be made in light of the above disclosure or may be acquired from practice of the Aspects. In addition, any of the Aspects described herein may be combined with any other Aspect unless explicitly stated otherwise. Features described in connection with a particular Aspect may be used interchangeably or in combination with features of any other Aspect, and the various implementations disclosed herein are not limited to the specific combinations expressly set forth in the Aspects above.

[0214] References

[0215] 1 . Souders CA, Bowers SL, Baudino TA. Cardiac fibroblast: the renaissance cell. Circ Res. 2009;105:1164-1 176. doi: 10.1161 / CIRCRESAHA.109.209809.

[0216] 2. Rubino M, Travers JG, Headrick AL, Enyart BT, Lemieux ME, Cavasin MA, Schwisow JA, Hardy EJ, Kaltenbacher KJ, Felisbino MB, et al. Inhibition of Eicosanoid Degradation Mitigates Fibrosis of the Heart. Circ Res. 2023;132:10-29. doi: 10.1 161 / CIRCRESAHA.122.321475.

[0217] 3. Nagaraju CK, Robinson EL, Abdesselem M, Trenson S, Dries E, Gilbert G, Janssens S, Van Cleemput J, Rega F, Meyns B, et al. Myofibroblast Phenotype and Reversibility of Fibrosis in Patients With End-Stage Heart Failure. J Am Coll Cardiol. 2019;73:2267-2282. doi: 10.1016 / j.jacc.2O19.02.049.

[0218] 4. Bray MA, Singh S, Han H, Davis CT, Borgeson B, Hartland C, Kost-Alimova M, Gustafsdottir SM, Gibson CC, Carpenter AE. Cell Painting, a high-content image based assay for morphological profiling using multiplexed fluorescent dyes. Nat Protoc. 2016;11 :1757-1774. doi: 10.1038 / nprot.2016.105.

[0219] 5. Travers JG, Tharp CA, Rubino M, McKinsey TA. Therapeutic targets for cardiac fibrosis: from old school to next-gen. J Clin Invest. 2022;132. doi: 10.1172 / JCI148554.

[0220] The following claims are intended to cover all of the generic and specific features of the invention herein described, and all statements of the scope of the invention that, as a matter of language, might be said to fall therebetween.

Claims

What is claimed is:1 . A computer-implemented method for analyzing cardiac fibroblasts, comprising: acquiring one or more multi-channel fluorescence images of stained cardiac fibroblasts; processing, by an image-processing module, the one or more multi-channel fluorescence images, the processing comprising: performing illumination correction on each fluorescence channel; segmenting nuclei, cytoplasm, and whole-cell regions within the images; extracting a plurality of morphological features from each cardiac fibroblast; filtering segmentation errors or debris; curating single-cell feature data; and normalizing the curated feature data to generate single-cell profiles; providing the single-cell profiles as input data to a machine-learning system configured to classify cardiac fibroblasts; and classifying, by the machine-learning system, one or more cardiac fibroblasts as failing or non-failing based on the input data.

2. The method of claim 1 , wherein the stained cardiac fibroblasts have been stained with an F-actin stain.

3. The method of claim 1 , wherein extracting the plurality of morphological features comprises extracting features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

4. The method of claim 1 , wherein extracting the plurality of morphological features comprises extracting features of at least nuclei and cytoplasm.

5. The method of claim 1 , wherein extracting the plurality of morphological features comprises extracting at least one of the following features: the mean intensity at the edge of Hoechst within the nucleus; the minimum intensity at the edge of Hoechst within the nucleus; the total integrated intensity of actin at the edge of the cytoplasm; and the maximum intensity at the edge of Hoechst within the cytoplasm.

6. The method of claim 1 , wherein curating the single-cell feature data comprises organizing the data into a tabular format.

7. The method of claim 1 , wherein the machine-learning system has been trained via a plurality of steps, including: receiving, by the machine-learning system, training input data comprising normalized single-cell profiles generated from the extracted morphological features of a plurality of cardiac fibroblasts; receiving, by the machine-learning system, corresponding ground-truth labels indicating whether each cardiac fibroblast in the training input data was obtained from failing or non-failing cardiac tissue; processing, by the machine-learning system, the training input data to generate predicted labels for the plurality of cardiac fibroblasts; comparing the predicted labels to the ground-truth labels to compute one or more evaluation metrics indicative of training performance; and adjusting, by the machine-learning system, one or more model parameters based on the evaluation metrics such that the predictive accuracy of the machine-learning system improves over successive training iterations.

8. The method of claim 1 , further comprising diagnosing heart failure in a subject based on the classification of the one or more cardiac fibroblasts as failing or non-failing.

9. The method of claim 1 , further comprising predicting a likelihood of heart failure in a subject based on the classification of the one or more cardiac fibroblasts as failing or non-failing.

10. The method of claim 1 , further comprising: acquiring first fluorescence images of cardiac fibroblasts obtained from a subject before administration of a therapeutic agent; acquiring second fluorescence images of cardiac fibroblasts obtained from the subject after administration of the therapeutic agent; processing, by the image-processing module, the first and second multi-channel fluorescence images to generate corresponding sets of normalized single-cell profiles, the processing comprising illumination correction, segmentation of nuclei and cytoplasm, extraction of morphological features, filtration of segmentation errors, curation of single-cell feature data, and normalization of the curated data comparing, by a computing device, morphological features extracted from the first images to corresponding morphological features extracted from the second images; anddetermining the efficacy of the therapeutic agent based on changes in one or more of the morphological features between the first and second time points.11 . The method of claim 1 , further comprising administering a therapeutic agent to a subject if the classification indicates that the subject has failing cardiac fibroblasts.

12. A computer-implemented method for predicting heart failure in a subject, comprising: processing, by an image-processing module, one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features extracted from the images; providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having heart failure or not having heart failure; and predicting, by the machine-learning system, whether the subject has heart failure based on the morphological features contained in the single-cell profiles.

13. The method of claim 12, wherein training the machine-learning system comprises receiving training input data comprising single-cell profiles of cardiac fibroblasts associated with clinical labels indicating heart failure status.

14. The method of claim 13, wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

15. The method of claim 12, wherein the morphological features used for prediction comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

16. The method of claim 12, wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their predictive importance for determining heart failure.

17. The method of claim 12, wherein predicting heart failure comprises generating a probability or numerical score that the subject has or will have heart failure.

18. The method of claim 12, wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

19. The method of claim 12, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

20. The method of claim 12, wherein the extracted morphological features comprise at least nuclei and cytoplasm.21 . A computer-implemented method for diagnosing heart failure in a subject, comprising: processing, by an image-processing module, one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features extracted from the images; providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having heart failure or not having heart failure; and diagnosing, by the machine-learning system, whether the subject has heart failure based on the morphological features contained in the single-cell profiles.

22. The method of claim 20, wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

23. The method of claim 21 , wherein training the machine-learning system comprises receiving training input data comprising single-cell profiles of cardiac fibroblasts associated with clinical labels indicating heart failure status.

24. The method of claim 21 , wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their diagnostic importance for determining heart failure.

25. The method of claim 21 , wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

26. The method of claim 21 , wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

27. The method of claim 21 , wherein the morphological features used for diagnosing heart failure comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

28. The method of claim 21 , wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

29. A computer-implemented method for screening the efficacy of a therapeutic agent, comprising: acquiring first multi-channel fluorescence images of cardiac fibroblasts obtained from a subject or sample prior to exposure to the therapeutic agent, and acquiring second multi-channel fluorescence images of cardiac fibroblasts obtained from the subject or sample following exposure to the therapeutic agent; processing, by an image-processing module, the first and second multi-channel fluorescence images to generate first and second sets of normalized single-cell profiles, the processing comprising performing illumination correction, segmenting nuclei and cytoplasm, extracting morphological features, filtering segmentation errors or debris, curating single-cell feature data, and normalizing the curated data; comparing, by a computing device, one or more morphological features from the first set of normalized single-cell profiles to corresponding morphological features from the second set of normalized single-cell profiles; and determining the efficacy of the therapeutic agent based on changes in the one or more morphological features between the first and second sets of normalized singlecell profiles.

30. The method of claim 29, wherein determining the efficacy of the therapeutic agent comprises determining whether one or more morphological features move toward values characteristic of non-failing cardiac fibroblasts.31 . The method of claim 29, wherein the morphological features compared comprise features extracted from at least nuclei and cytoplasm.

32. The method of claim 29, wherein determining the efficacy of the therapeutic agent comprises determining whether a morphological feature increases, decreases, falls within a predetermined range, or exceeds a predetermined threshold.

33. The method of claim 29, wherein comparing the morphological features comprises comparing one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

34. The method of claim 29, further comprising providing the first and second sets of normalized single-cell profiles to a machine-learning system configured to compute a metric indicative of phenotypic change induced by the therapeutic agent.

35. The method of claim 29, wherein the therapeutic agent is determined to be effective when the machine-learning system computes a metric indicating that the second set of normalized single-cell profiles is more similar to profiles of non-failing cardiac fibroblasts than to profiles of failing cardiac fibroblasts.

36. The method of claim 29, wherein determining the efficacy of the therapeutic agent comprises applying one or more rule-based, threshold-based, or algorithmic criteria to the morphological feature changes.

37. The method of claim 29, further comprising acquiring additional fluorescence images of cardiac fibroblasts at one or more intermediate or later time points following exposure to the therapeutic agent.

38. A method of treating a subject, comprising: acquiring one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject; processing, by an image-processing module, the images to generate single-cell profiles comprising morphological features extracted from the cardiac fibroblasts; classifying, by a machine-learning system or by one or more rule-based criteria, the cardiac fibroblasts as failing or non-failing based on the morphological features; and administering a therapeutic agent to the subject when the cardiac fibroblasts are classified as failing.

39. The method of claim 38, wherein classifying the cardiac fibroblasts comprises providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles associated with clinical labels indicating heart failure status.

40. The method of claim 38, wherein classifying the cardiac fibroblasts comprises determining whether one or more morphological features move toward values characteristic of failing cardiac fibroblasts.41 . The method of claim 38, wherein classifying the cardiac fibroblasts comprises evaluating one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

42. The method of claim 38, wherein the therapeutic agent comprises a treatment indicated for heart failure, fibrosis, or cardiac remodeling.

43. The method of claim 38, further comprising acquiring additional cardiac fibroblasts from the subject after administration of the therapeutic agent andevaluating changes in the morphological features to determine whether the treatment is effective.

44. The method of claim 43, further comprising modifying the therapeutic agent or its dosage based on changes in the morphological features following treatment.

45. The method of claim 38, wherein the therapeutic agent is administered when at least one morphological feature exceeds or falls below a predetermined cutoff.

46. The method of claim 38, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

47. The method of claim 38, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

48. A system for analyzing cardiac fibroblasts, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: acquire one or more multi-channel fluorescence images of stained cardiac fibroblasts; perform illumination correction on each fluorescence channel; segment nuclei, cytoplasm, and whole-cell regions within the images; extract a plurality of morphological features from each cardiac fibroblast; filter segmentation errors or debris; curate single-cell feature data; normalize the curated feature data to generate single-cell profiles; and provide the single-cell profiles to a machine-learning system configured to classify one or more cardiac fibroblasts as failing or non-failing based on the single-cell profiles.

49. The system of claim 48, wherein the stained cardiac fibroblasts have been stained with an F-actin stain.

50. The system of claim 48, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.51 . The system of claim 48, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

52. The system of claim 48, wherein the extracted morphological features comprise at least one of: the mean intensity at the edge of Hoechst within the nucleus; the minimum intensity at the edge of Hoechst within the nucleus; the total integrated intensity of actin at the edge of the cytoplasm; and the maximum intensity at the edge of Hoechst within the cytoplasm.

53. The system of claim 48, wherein curating the single-cell feature data comprises organizing the data into a tabular format.

54. The system of claim 48, wherein the machine-learning system has been trained via operations comprising: receiving training input data comprising normalized single-cell profiles generated from morphological features extracted from a plurality of cardiac fibroblasts; receiving corresponding ground-truth labels indicating whether each cardiac fibroblast was obtained from failing or non-failing cardiac tissue; processing the training input data to generate predicted labels; comparing the predicted labels to the ground-truth labels to compute one or more evaluation metrics; and adjusting model parameters based on the evaluation metrics.

55. The system of claim 48, further configured to diagnose heart failure in a subject based on the classification of the cardiac fibroblasts as failing or non-failing.

56. The system of claim 48, further configured to predict a likelihood of heart failure in a subject based on the classification of the cardiac fibroblasts as failing or non-failing.

57. The system of claim 48, further configured to: acquire first fluorescence images of cardiac fibroblasts obtained from a subject before administration of a therapeutic agent and second fluorescence images after administration of the therapeutic agent; process the first and second fluorescence images to generate corresponding sets of normalized single-cell profiles; compare morphological features extracted from the first images to corresponding morphological features extracted from the second images; and determine an efficacy of the therapeutic agent based on changes in one or more morphological features.

58. The system of claim 48, further configured to recommend administration of a therapeutic agent to a subject when the classification indicates that the subject has failing cardiac fibroblasts.

59. A system for diagnosing heart failure in a subject, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process one or more multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features; and provide the single-cell profiles to a machine-learning system trained using single-cell profiles extracted from cardiac fibroblasts obtained from subjects clinically characterized as having or not having heart failure, the machinelearning system being configured to diagnose whether the subject has heart failure based on the morphological features.

60. The system of claim 59, wherein the machine-learning system is trained using training input data comprising single-cell profiles associated with clinical labels indicating heart failure status.61 . The system of claim 59, wherein the machine-learning system assigns weighting coefficients to one or more of the extracted morphological features based on their diagnostic importance for determining heart failure.

62. The system of claim 59, wherein training further comprises generating predicted clinical labels for the training input data, comparing the predicted labels to the clinical labels, and adjusting model parameters based on one or more evaluation metrics.

63. The system of claim 59, wherein the extracted morphological features comprise features of at least nuclei and cytoplasm.

64. The system of claim 59, wherein the extracted morphological features comprise features of nuclei, cytoplasm, mitochondria, endoplasmic reticulum, plasma membrane, and Golgi apparatus.

65. The system of claim 59, wherein the morphological features used for diagnosing heart failure comprise one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

66. The system of claim 59, wherein the images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

67. A system for screening the efficacy of a therapeutic agent, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process first and second multi-channel fluorescence images of cardiac fibroblasts obtained before and after exposure to a therapeutic agent to generate first and second sets of normalized single-cell profiles; compare one or more morphological features from the first set of normalized single-cell profiles to corresponding morphological features from the second set of normalized single-cell profiles; and determine the efficacy of the therapeutic agent based on changes in the one or more morphological features between the first and second sets of normalized single-cell profiles.

68. The system of claim 67, wherein determining the efficacy of the therapeutic agent comprises determining whether one or more morphological features move toward values characteristic of non-failing cardiac fibroblasts.

69. The system of claim 67, wherein the morphological features compared comprise features extracted from at least nuclei and cytoplasm.

70. The system of claim 67, wherein determining the efficacy of the therapeutic agent comprises determining whether a morphological feature increases, decreases, falls within a predetermined range, or exceeds a predetermined threshold.71 . The system of claim 67, wherein comparing the morphological features comprises comparing one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

72. The system of claim 67, further configured to provide the first and second sets of normalized single-cell profiles to a machine-learning system configured to compute a metric indicative of phenotypic change induced by the therapeutic agent.

73. The system of claim 72, wherein the metric comprises a score, probability, similarity value, or reduced-dimensionality representation reflecting morphological differences between the first and second sets of normalized single-cell profiles.

74. The system of claim 72, wherein the therapeutic agent is determined to be effective when the machine-learning system computes a metric indicating that the second set of normalized single-cell profiles is more similar to profiles of non-failing cardiac fibroblasts than to profiles of failing cardiac fibroblasts.

75. The system of claim 67, wherein determining the efficacy of the therapeutic agent comprises applying one or more rule-based, threshold-based, or algorithmic criteria to the morphological feature changes.

76. The system of claim 67, further configured to process fluorescence images acquired at one or more intermediate or later time points following exposure to the therapeutic agent.

77. A system for supporting treatment of a subject, comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to: process multi-channel fluorescence images of cardiac fibroblasts obtained from the subject to generate single-cell profiles comprising morphological features; classify the cardiac fibroblasts as failing or non-failing based on the morphological features; and determine, via a treatment-decision module, whether a therapeutic agent should be administered to the subject based on the classification.

78. The system of claim 77, wherein classifying the cardiac fibroblasts comprises providing the single-cell profiles to a machine-learning system trained using training data comprising single-cell profiles associated with clinical labels indicating heart failure status.

79. The system of claim 77, wherein classifying the cardiac fibroblasts comprises determining whether one or more morphological features move toward values characteristic of failing cardiac fibroblasts.

80. The system of claim 77, wherein classifying the cardiac fibroblasts comprises evaluating features extracted from at least nuclei and cytoplasm.81 . The system of claim 77, wherein classifying the cardiac fibroblasts comprises evaluating one or more of the mean intensity at the edge of Hoechst within the nucleus, the minimum intensity at the edge of Hoechst within the nucleus, the total integrated intensity of actin at the edge of the cytoplasm, and the maximum intensity at the edge of Hoechst within the cytoplasm.

82. The system of claim 77, wherein the therapeutic agent recommended for administration comprises a treatment indicated for heart failure, fibrosis, or cardiac remodeling.

83. The system of claim 77, wherein the cardiac fibroblasts comprise human cardiac fibroblasts derived from failing and non-failing human hearts.

84. The system of claim 77, wherein the multi-channel fluorescence images of cardiac fibroblasts include cardiac fibroblasts that have been stained with an F-actin stain.

Images