The development of an artificial intelligence software utilising a variety of AI techniques has significantly enhanced and expanded laboratory research into myocardial contractility and the management of calcium levels, essential for its proper function.

World Heart Report 2023, issued by the World Heart Federation in Geneva, regrettably reaffirms a well-known grim reality: cardiovascular diseases still lead as the global cause of death for both genders.

The report outlines that over half a billion individuals are affected by these diseases, with around 60 million new cases annually. The peak in 2021 saw 20.5 million deaths, representing nearly a third of all global fatalities, with an average of 56,000 deaths per day, or one every 1.5 seconds.

These statistics underscore the critical importance of ongoing, updated, and thorough research into cardiac function, the pathologies leading to its decline, the roles of genetic or multifactorial diseases, and the efficacy of drugs in clinical trials.

Currently, laboratory studies on cardiac function lack a unified tool capable of swiftly and precisely processing video data from all available in vitro cardiac models, whilst concurrently assessing their contraction capabilities and calcium levels.
A team of biomedical engineers at Columbia University in New York has bridged this gap by integrating machine learning and deep learning techniques within a single system, validated in experiments for identifying cardiac diseases and categorizing the reactions of in vitro models to various cardiovascular drugs.
Looking ahead, the methodologies developed by researchers in the USA could empower clinicians to undertake preventative measures through progressively earlier and more accurate diagnoses of cardiac diseases – the primary cause of mortality globally – thereby enhancing the quality of patient care delivered.

In vitro study methodologies of cardiac function

Research on human heart function is currently conducted in laboratories through “in vitro cardiac modelling.” This involves the analysis of clinical images of myocardial models, which range from one-dimensional (such as isolated cardiac muscle tissue cells derived from human stem cells), to two-dimensional (comprising single layers or small groups of cardiac muscle tissue, also derived from human stem cells), and up to three-dimensional multicellular contractile tissues.

These tissues are studied either in vitro or “in vivo” (i.e., outside a test tube), and include cardiac spheroids (three-dimensional cell aggregates that mimic heart characteristics), engineered cardiac tissues, and hearts from animals and humans donated post-mortem for scientific research.

The primary focus lies on two critical aspects of cardiac function: contractility (the heart’s capacity to contract) and the management of calcium levels [source: “In Vitro Methods to Model Cardiac Mechanobiology in Health and Disease” – National Library of Medicine, 2021].

Particularly concerning calcium, the Italian Journal of Cardiology underscores its pivotal role in myocardial contraction, but also highlights that the formation of calcium deposits in the arteries (calcification) is frequently associated with severe and potentially fatal vascular damage. This remains «a significant challenge for interventional cardiologists». This underpins the importance and necessity of evaluating, in vitro, the dynamics of contractions alongside the management of blood calcium levels.

In vitro study of cardiac function: limitations

In terms of analysing heart function using in vitro cardiac models, the “BeatProfiler: Multimodal In Vitro Analysis of Cardiac Function Enables Machine Learning Classification of Diseases and Drugs” (IEEE Open Journal of Engineering in Medicine and Biology, April 5, 2024), authored by the Department of Biomedical Engineering at Columbia University in New York, details that this analysis is conducted by «extracting traces of cardiac contractions and calcium signals from a series of model videos, using bright field and contrast imaging with ion-sensitive dyes». After initially evaluating these traces:

«… it is necessary to meticulously process the videos related to contractility and calcium levels to derive metrics that accurately reflect changes observed in the functions of the in vitro models following the induction of a specific pathology or drug exposure»

At this stage, the Columbia University team is notably critical, pointing out the current absence of a single, standardized tool that is universally accessible and capable of precisely processing clinical imaging data from a diverse array of in vitro myocardial models, from the simplest one-dimensional models to the most complex three-dimensional tissues, including engineered, animal, and human models.

«External programs such as ImageJ or MATLAB are frequently employed, necessitating manual input of parameters, which complicates high-throughput analysis. Additionally, many methods are susceptible to noise and artefacts, which can impede accurate low-signal video analysis», the researchers state. They further lament «the lack of a single software solution capable of concurrently analysing images related to both contractile functions and calcium level management».

Highlighting an example of the limited range of myocardial models considered in previous research, the team refers to a Japanese study from 2014 – “Image-based evaluation of contraction-relaxation kinetics of human-induced pluripotent stem cell-derived cardiomyocytes: Correlation and complementarity with extracellular electrophysiology“, National Library of Medicine – which focused solely on contractile function in one-dimensional heart models, thereby limiting the breadth and authority of the research.

Furthermore, a joint study from 2018 by British and Dutch universities, “MUSCLEMOTION: A Versatile Open Software Tool to Quantify Cardiomyocyte and Cardiac Muscle Contraction In Vitro and In Vivo“, National Library of Medicine, commendably introduced a unified open-source method that automates image processing across multiple in vitro myocardial models. However, it solely concentrated on analyzing heart contraction, overlooking the management of normal calcium levels.

Machine learning-based approaches in the in vitro study of cardiac function

In the realm of automating data processing for video data on cardiac function observed across various in vitro myocardial models, numerous studies have utilised machine learning techniques to classify the heart’s contractile responses based on varying levels of blood calcium, involving both healthy and diseased cardiac muscle cells.

However, as noted by biomedical engineers from New York University, these studies predominantly used algorithms trained with datasets based on predefined myocardial functionality parameters. These parameters are sufficient for simpler processing tasks, such as classifying video data related to well-known diseases with distinct characteristics and straightforward diagnoses.

Nevertheless, for processing more complex clinical imaging data, such as those concerning lesser-known and more insidious pathologies or feedback on specific types of cardiovascular drugs under trial, these parameters prove inadequate due to their selection from limited and narrowly focused datasets.

It is deep learning – a subset of machine learning – that, according to researchers, could potentially overcome this limitation, «although it has thus far only been applied to binary classifications of in vitro cardiac models derived from stem cells».

AI techniques for classifying responses to cardiovascular drugs

With the objective of transcending mere binary data classification and the use of one-dimensional and two-dimensional in vitro cardiac models, typical of previous studies on video data processing of cardiac function, the authors of the aforementioned New York study – published in the Open Journal of Engineering in Medicine and Biology on April 5, 2024 – have developed an artificial intelligence software (BeatProfiler) based on machine learning and deep learning algorithms for the rapid processing of large datasets concerning myocardial contractility and calcium level management, originating from one-dimensional, two-dimensional, and three-dimensional cardiac models, including engineered, animal, and human models.

This open-source AI software, freely available for direct use by any laboratory, was tested by the researchers in a classification trial of heart responses to four commonly marketed cardiovascular drugs (Quinidine, an antiarrhythmic; Propranolol, a beta-blocker; E-4031, an experimental antiarrhythmic; and Verapamil, a vasodilator), based on their mechanisms of action on in vitro cardiac models, which had previously induced restrictive cardiomyopathy, leading to stiffness of the heart walls and thus impeding proper contraction and filling.

A significant amount of data concerning restrictive myocardial pathology and in vitro cardiac modelling was sourced from a prior study by the same team from Columbia University in New York [“Engineered cardiac tissue model of restrictive cardiomyopathy for drug discovery” – Cell Reports Medicine, 2023], specifically focused on the development of an engineered cardiac tissue model for new drug studies targeting restrictive cardiomyopathy. These data were employed to train the machine learning algorithms tasked with processing clinical imaging data derived from the in vitro cardiac models, for testing disease identification and classification (in this case, restrictive cardiomyopathy).

Regarding the data used to train the deep learning algorithms for classifying responses to the four selected cardiovascular drugs for the test, the final dataset included 12,428 myocardial beat types in response to Quinidine10,759 to Propranolol12,383 to E-4031, and 11,696 to Verapamil.

The authors note that during the experiment, fifteen clinical videos were selected for examination (five for each cardiac in vitro model: monolayers, engineered tissues, and cardiac spheroids) for each contractile mode and calcium level, and the drugs were administered to each tissue one hour before video acquisition, which was conducted through an sCMOS camera connected to a microscope.

The study featured a comparative analysis, contrasting the test results with those from a control group of cardiac models to which no drugs had been administered. This analysis revealed that the AI software from the team of biomedical engineers provided more accurate detection of video signals of contraction and calcium levels (even in cases of low signal), reduced false positives, and increased analysis speed from 7 to 50 times.

In detail, the features extracted from the videos by the software classified the cardiac function in the presence of restrictive cardiomyopathy with 98% accuracy.

Glimpses of Futures

The research presented validates the role of artificial intelligence techniques in supporting the comprehensive and detailed in vitro analysis of cardiac function and responses to cardiovascular drugs across diverse heart models. This opens up scenarios that were unimaginable a decade ago, when the field was limited to analyzing simplistic unidimensional cardiac models. These models were inadequate for fully understanding the complexity of human heart physiology and the intricacies of factors contributing to its decline.

Let us now consider potential future scenarios, analysing, via the STEPS matrix, the impacts that the advancement of data processing techniques utilising large video datasets on myocardial contractility and calcium levels in the blood vessels – from unidimensional, bidimensional, and tridimensional in vitro heart models, including engineered, animal, and human models – may have across various aspects.

S – SOCIAL: the potential of open-source AI software to advance in vitro studies on cardiac biology could enable quicker identification and classification of heart diseases by researchers and more precise evaluations of potential pharmacological therapies. The primary goal is to support cardiologists in making early and accurate diagnoses and developing increasingly specific and targeted treatments, which is «the principal objective of both scientific research and clinical practice, » as outlined in the document by the “Consultation of Scientific Societies for Cardiovascular Risk Reduction“. This would facilitate the detection of minimal cardiovascular damage and the earliest signs of decline, allowing for immediate risk assessment for the organ. This has a positive impact socially and in clinical therapy, offering opportunities for immediate intervention for patients, aiming to improve medical and healthcare services for those affected by diseases that, in our country alone, account for 35.8% of all deaths and over 900,000 hospital admissions annually [source: “Cardiomyopathies Matter. A Roadmap for Improving the Diagnosis and Care of Patients with Cardiomyopathy in Italy“].

T – TECHNOLOGICAL: in the future, the artificial intelligence techniques employed by the Columbia University team for processing large datasets of video data on myocardial contractility and calcium levels must be refined. To broaden and generalize their use across a diverse array of cardiovascular diseases and for classifying effects on the myocardium by different drug classes, the machine learning and deep learning algorithms need enhancement. The long-term objective is to tailor the developed AI software to the pharmaceutical context of the reference country, to accelerate the testing of all contemporary cardiac drug candidates. «A limitation of our study is that we tested the classification of a restricted set of drugs as a proof of concept. A broader spectrum of drugs should instead be incorporated into the dataset to validate the deep learning model’s generalizability in future studies», the authors note, who are already striving to broaden their AI system’s capabilities for new applications in in vitro cardiac function research, including comprehensive analyses of disorders affecting the dynamics of blood pumping by the atria and ventricles.

E – ECONOMIC: The European House – Ambrosetti (TEHA) has noted the significant economic impact on our healthcare system of cardiovascular, cerebral, and vascular diseases – in Italy, these are the leading causes of death, hospitalization, and among the main reasons for disability – «estimated at approximately 19-24 billion euros as of March 2023. This sum includes 14-16 billion euros in direct healthcare costs (80% attributed to hospitalization costs) and 5-8 billion euros in indirect costs, both healthcare and non-healthcare, stemming from the loss of productivity of the patient and caregiver of working age, as well as social security and welfare expenses». In this rather alarming scenario, a potential future where automated in vitro analysis of cardiac function and response to cardiovascular drugs – facilitated by freely available, open-source artificial intelligence software – could enable increasingly early and accurate diagnoses of heart disorders, contributing to a long-term reduction of all those socially adverse effects caused by late-diagnosed cardiovascular, cerebral, and vascular diseases, thereby potentially reducing healthcare costs.

P – POLITICAL: for future prevention to rest on robust diagnostics, immediately followed by risk quantification and prompt patient intervention – as envisaged by the developers of the described AI software – there is an urgent need for a strategic “plan.” This necessity was highlighted by The European House – Ambrosetti (TEHA) when, in December 2023, it advocated for the establishment of a working group by the Ministry of Health, aimed at formulating a National Plan for Cardiovascular, Cerebral, and Vascular Diseases. «For too long, these high-impact diseases have lingered on the margins of the health agenda: to date, they are the only major disease group not covered by a dedicated National Plan, unlike in other European countries», TEHA analysts observed. Notably, between 1990 and 2020, access to technological and pharmacological innovation in Italy led to a reduction in the mortality rate for these conditions. However, significant challenges remain that must be addressed if we are to fully benefit from advances in in vitro studies on cardiac function enabled by artificial intelligence.

S – SUSTAINABILITY: the prevention of cardiovascular diseases through timely diagnostics and immediate interventions on patients, structured and organized under a National Plan, must be equitable. The progress in in vitro studies on cardiac biology, enabling faster identification and classification of myocardial pathologies, is negated if access to care is limited. In this context, in November 2023, the four-year EU project named JACARDI(an acronym for Joint Action on CARdiovascular diseases and DIabetes) was launched, concerning cardiovascular diseases and diabetes, for which the European Commission allocated a significant 53 million euros, appointing our National Institute of Health (ISS) as the European coordinator and leader. A distinctive feature of the initiative is its focus on addressing the health challenges (present and future) posed by the two diseases inclusively, «with a specific focus on the social determinants of health, cultural and ethnic diversity among patients, and the promotion of equity, including from a gender perspective, » identifying the main social dimensions of disparities in the two diseases.

Written by: