Artificial Intelligence in Cardiac Disease Diagnosis: A Comprehensive

Investigation

Tianhao Zhang

Robot Engineering, Shenzhen Technology University, Shenzhen, China

Keywords: Artificial Intelligence, Cardiac Disease, Machine Learning.

Abstract: In the medical field, the use of massive data to assist medical diagnosis is an inevitable trend of development.

In the diagnostic process, various machine learning algorithms are utilized to achieve assisted medical

diagnosis of cardiac diseases based on a large amount of data sources acquired in clinical practice. This paper

introduces the role of artificial intelligence (AI) in the diagnosis of cardiac diseases, and describes the

utilization of various traditional machine learning and deep learning models to improve diagnostic efficiency

and accuracy. By examining large amounts of clinical data, including electronic health records and imaging,

AI has a unique advantage over traditional diagnostic methods in terms of high accuracy and efficiency. This

paper explores a variety of AI diagnostic frameworks. In addition, this paper explores the limitations and

challenges faced by AI in the field of medical diagnostics today, including issues of data quality, model

interpretability, and population generalization, and also proposes corresponding approaches such as federated

learning and Explainable AI are also proposed as possible solutions to overcome these obstacles. This paper

not only demonstrates the current progress of AI in the field of cardiac diagnosis, but also makes predictions

about its future prospects.

1 INTRODUCTION

Heart disease is the disease that the causes are

attributed to structure or function of the heart,

including myocardial infarction, arrhythmias, heart

failure, and various other types. It has become one of

the main causes of disability and death worldwide

with the trend of population aging and the change of

lifestyle. Millions of people die each year due to heart

disease, highlighting the importance of diagnosing

heart disease.

Relying on the experience of physicians and a

series of examinations such as electrocardiograms

and echocardiograms, traditional medical diagnostic

methods have certain limitations, such as slow

diagnostic process, misdiagnose, and various other

types. In addition, these methods incur high labor

costs. Consequently, an increasing number of

researchers tend to explore approaches with Artificial

Intelligence (AI) to improve the diagnostic methods

since it can effectively discover the correlations

within vast amounts of data by mimicking human

https://orcid.org/0009-0006-9812-4889

learning and reasoning processes (Qiu, 2019; Qiu,

2022).

AI is widely applied to many fields including

healthcare (Li, 2024; Liu, 2023; Zhao, 2023). And

this transformative shift brings about a new era in

medical diagnosis, patient care, and treatment

personalization. A systematic review by Kakas et al.

underscores the latest developments in Explainable

artificial intelligence (XAI) solutions for medical

decision support, emphasizing the necessity of

increased collaboration between medical and AI

experts to devise frameworks guiding the design,

implementation, and evaluation of XAI solutions in

medicine (Prentzas, 2023). In addition, research by

Feng et al. discusses common applications of AI in

medicine, including virtual assistants, AI-augmented

diagnostics, and medical robots (Zhao, 2022).

Moreover, High-Performance Computing (HPC)

technologies show great potential in processing the

vast amounts of data characteristic of modern medical

practice (Koch, 2023).

In addition to these applications, AI is widely used

in the field of cardiac diagnosis. For example, Butler

Zhang, T.

Artiﬁcial Intelligence in Cardiac Disease Diagnosis: A Comprehensive Investigation.

DOI: 10.5220/0012911200004508

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 135-140

ISBN: 978-989-758-713-9

135

et al. developed Electrocardiography Artificial

Intelligence (ECG-AI) models for predicting fatal

coronary heart disease using 12-lead and single-lead

ECG data, demonstrating the potential of AI for early

detection and intervention (Butler, 2023). Similarly,

Nedadur et al. explored the use of AI in

echocardiography, particularly in valvular heart

disease, demonstrating how AI can improve

diagnostic accuracy through automated analysis

(Nedadur, 2022). Maambo et al. described an AI

medical diagnostic system that utilizes healthcare

data to predict heart disease risk, using data mining

algorithms to improve the diagnostic process

(Maambo, 2022). Janik et al. investigate deep

learning models for cardiac Magnetic Resonance

Imaging (MRI) segmentation, focusing on the

importance of model interpretability to make AI-

assisted diagnosis more transparent and

understandable (Janik, 2021). Finally, Thoenes et al.

discuss the use of AI in the management of aortic

valve disease, emphasizing how AI can support early

diagnosis and efficient treatment planning (Thoenes,

2021). Together, these examples highlight the diverse

and impactful applications of AI in the diagnosis,

treatment, and management of heart disease, marking

an important step in cardiology's use of technology to

improve patient prognosis.

The review is divided into following sections. The

second section firstly provides an overview of current

mainstream methods for cardiac disease diagnosis.

Following this is an analysis of those methods,

including their advantages and disadvantages, the

difficulties they encounter and the main algorithms

they used. The last section is conclusion of the current

AI field of the disease diagnosis, and its future

prospects.

2 METHOD

In this section, the current mainstream AI-based

models for diagnosing heart disease, including

traditional machine learning, as well as deep learning

models, will be demonstrated.

2.1 Framework of AI-Based Heart

Disease Diagnosis

In order to improve the accuracy of diagnosing heart

disease as well as the efficiency of the diagnosis and

to improve the predictive ability, an AI-based

diagnostic framework for heart disease should

contain multiple steps. Typically, a typical

framework is going to contain the following aspects.

The first step is data collection and preprocessing.

The sources of data are generally large datasets

including Electronic Health Records (EHR), imaging

data (e.g., echocardiograms, MRIs, Computed

Tomography (CT) scans), genomics, wearable

devices, and patient-reported data. However,

generally these data are generally not directly

available as inputs to the training model, and before

that, they need to be cleaned, normalized and

anonymized so that the model can recognize them

effectively. This step may include dealing with

missing values, correcting errors and normalizing

formats. Next, the model needs to extract the most

relevant features that will help to accurately diagnose

heart disease (e.g., patient demographics, clinical

parameters, lab results, imaging features) and will

convert the raw data into a format that can be

efficiently processed by AI algorithms, often using

techniques such as Principal Component Analysis

(PCA) for dimensionality reduction or customized

algorithms for extracting meaningful attributes from

complex data such as images. In the third step,

appropriate artificial intelligence models such as

Machine Learning (ML) algorithms (e.g., decision

trees, support vector machines, random forests) and

Deep Learning (DL) models (e.g., convolutional

neural networks for image analysis) are selected

based on the characteristics of the data. Preprocessed

data is fed into selected models to learn patterns

related to heart disease. This involves dividing the

data into training and validation sets to iteratively

improve the accuracy of the model. In order to verify

the validity of the model, it needs to be validated and

tested after the training is completed. Usually, this

step involves tuning the model parameters using a

separate part of the dataset (in the dataset, but not

involved in model training). The performance of the

model on the test set is then evaluated to assess its

generalizability and accuracy in diagnosing heart

disease. Performance metrics may include accuracy,

sensitivity, specificity, and area under the receiver

operating characteristic curve (ROC).

For this type of AI models, there is also a need to

integrate them into the clinical workflow by

integrating the AI models into the clinical

environment, for example by embedding them in

electronic medical record systems or diagnostic tools

to support healthcare professionals. User-friendly

interfaces also need to be developed to enable

clinicians to enter data, receive predictions and

visualize results in an intuitive way.

To ensure that it remains accurate and effective

over time, the developers involved also have to

continuously monitor the performance of the AI

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

136

system in the real world. They also need to

incorporate feedback and outcome data from

clinicians to refine and update the model and ensure

that it adapts to new data and changing clinical

practices.

2.2 Traditional Machine

Learning-Based Heart Disease

Diagnosis

2.2.1 Random Forest Classifier

The Random Forest classifier operates by

constructing a multitude of decision trees at training

time and outputting the class that is the mode of the

classes of the individual trees. A key aspect of this

method is its use of "bagging" (bootstrap aggregating)

to promote variance reduction across trees without

increasing bias. This is achieved by randomly

selecting subsets of the training data, with

replacement, to build each tree, ensuring diversity.

The Random Forest then utilizes feature importance

scores derived from the trees to identify and select the

most relevant features for heart disease prediction.

And specifically in the field of heart disease

diagnosis, Saranya et al. build on this by tuning the

hyperparameter through grid search approach and

successfully increasing the diagnostic accuracy to

96.53%. The advantage of this approach is that it is

able to handle large datasets with high dimensionality

and inherent feature selection capabilities, making it

well suited for complex health datasets (Saranya,

2023).

2.2.2 Support Vector Machine (Svm)

Support Vector Machine works by mapping input

features into high-dimensional feature spaces where

it becomes easier to linearly separate data points

belonging to different classes. The core principle of

SVM is to find the optimal separating hyperplane that

maximizes the margin between the closest points of

different classes, known as support vectors. This

method is particularly effective for heart disease

diagnosis due to its ability to deal with non-linear

boundaries through the use of kernel functions (e.g.,

linear, polynomial, and radial basis function). Suresh

et al. demonstrated a hybrid approach that combines

the feature selection strength of Random Forest with

the classification power of SVM, achieving an

accuracy of 98.3%. The hybrid model benefits from

SVM's robustness against overfitting in high-

dimensional spaces and its capacity for handling

nonlinear relationships, crucial for accurately

classifying medical data (Suresh, 2022).

2.2.3 Hybrid Random Forest with a Linear

Model

In the exploration of deep learning-based methods for

heart disease diagnosis, a notable advancement was

introduced by Dwarakanath B. and colleagues, who

proposed a novel approach that combines feature

selection with hybrid deep learning for heart disease

detection and classification (FSHDL-HDDC). This

method utilizes an Attention-based Convolutional

Neural Network (ACNN) combined with Long Short-

Term Memory (LSTM) specifically for analysing

medical data. By employing a feature selection

method based on the Elite Opposition-based Squirrel

Search Algorithm (EO-SSA), it identifies the optimal

subset of features. This approach has demonstrated

exceptionally high accuracy in the field of heart

disease diagnosis, achieving a maximum accuracy

level of 97.72%. Its strength lies in its ability to

deeply analyse complex patterns in medical data,

thereby improving the precision and efficiency of

heart disease predictions, making it highly suitable

for handling health datasets with high dimensionality

and complexity (Mohan, 2019).

2.3 Deep Learning-Based Heart

Disease Diagnosis

This year has seen breakthroughs in the field of deep-

based learning. Deep learning has unique advantages

over traditional machine learning. First, deep learning

models typically have more parameters and

hierarchies, which means they can learn

automatically and extract more complex, abstract

features from data. Second, deep learning models

have more powerful representations, which enables

them to capture complex connections in large datasets

and thus achieve more accurate predictions. In

addition, thanks to back-propagation algorithms,

deep learning models can automatically tune model

parameters to optimize performance, and this end-to-

end training approach makes model construction and

tuning more efficient. Nevertheless, deep learning

models face problems such as high consumption of

computational resources and the training process is so

abstract that it is difficult to observe compared with

traditional machine learning models. Three effective

deep learning models will be described below

(Dwarakanath, 2022).

Artiﬁcial Intelligence in Cardiac Disease Diagnosis: A Comprehensive Investigation

137

2.3.1 Hybrid Deep Learning for Heart

Disease Detection and Classification

This model combines attention-based convolutional

neural network (ACNN) with long short-term

memory (LSTM) to analyse medical data efficiently

and accurately. The key to this approach is a special

selection method that identifies the best subset of

features based on elite opposition-based squirrel

search algorithm (EO-SSA). This method is proposed

by Dwarakanath B. et al. for heart disease detection

and classification (FSHDL-HDDC) in eHealth

environment. With this method, heart disease can be

predicted with up to 97.72% accuracy. This

demonstrates the potential of combining FEATURE

SELECTION algorithms and DEEP LEARNING

models (Kusuma, 2022).

2.3.2 BiDLNet: Integrated Model for

ECG-Based Diagnosis

The BiDLNet model is an integrated deep learning

framework proposed by Kusuma et al. , which utilizes

capabilities using electrocardiogram (ECG) data to

improve diagnostic capabilities and predictive

accuracy. BiDLNet employs a discrete wavelet

transform to extract two levels of features from the

data, followed by an ensemble classification scheme

that combines predictions from various deep learning

models. In terms of achievements, the model

achieved excellent results in dichotomization and

multichotomization of heart disease, 97.5% and

91.5%, respectively. The method exemplifies the

effectiveness of integrating multiple deep learning

techniques to improve the accuracy and reliability of

diagnosing heart disease from ECG signals (Hamad,

2021).

2.3.3 Long Short-Term Memory (LSTM)

Networks for Feature Extraction

LSTM networks are designed to avoid the long-term

dependency problem typical of standard recurrent

neural networks, making them adept at processing

sequences of data for tasks like heart sound signal

analysis. These networks introduce memory cells that

can maintain information in memory for long periods

of time. Each cell decides through structures called

gates (input, output, and forget gates) whether to

retain or discard information based on the strength

and relevance of incoming signals. Guven et al.

Guven and Uysal leveraged this capability by

combining both short-term features, extracted from

five-second heart sound fragments, and long-term

features from the entire signal.This approach can

combine the nuances of individual pieces of

information with the overall pattern of information,

benefiting from the ability of LSTM to capture and

learn from these complex temporal dependencies, so

that the diagnostic accuracy of heart disease can be

greatly enhanced (Guven, 2023).

3 DISCUSSION

Despite the breakthroughs in the field of diagnosing

heart disease, AI is still limited and faces many

challenges. Here are some of the major limitations

and challenges, as well as possible solutions.

3.1 Challenges and Limitations

3.1.1 Data Quality and Availability

Being able to guarantee the availability of high-

quality and annotated medical datasets is a major

challenge in the application of machine learning and

deep learning models. Because, Cardiac imaging and

EHR often contain sensitive patient information,

which is likely to raise privacy and ethical concerns.

In addition, the quality of residual datasets and

incomplete records can hinder the training of models

and affect the accuracy of diagnosis. These data

issues are the current need to be urgently addressed.

3.1.2 Model Interpretability

A long-standing difficulty in the field of deep

learning models is the "black box" problem, where

developers cannot know the decision-making process

of a model because of the abstract nature of the

model, which makes it difficult to trust the predicted

results to clinicians and patients, and this lack of

interpretability hinders the application of AI tools in

clinical practice. Although there are currently some

visualizations available that allow the detection of the

model's decision-making process, the "black box" is

still a major problem facing deep learning models

today.

3.1.3 Generalization Across Diverse

Populations

Inevitably, bias is present in training datasets, which

makes it difficult for machine learning models and

deep learning models to generalize their scope of

action to different populations. Models trained on a

specific population dataset may not perform well on

data from other races or age groups, which may lead

to inaccurate diagnoses.

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

138

3.2 Future Prospects and Possible

Solutions

3.2.1 Federated Learning and Data Privacy

Federated learning is known to be a promising

solution to the problem of data privacy and residual

data quality. The most important mention of federated

learning is decentralization. The general principle of

the model is that it takes its dataset from multiple

decentralized devices and servers and trains the

model on these devices simultaneously, which means

that the data is all stored locally on the devices

without the need to centralize the data centrally for

training. This approach not only improves privacy,

but also creates robust models that can learn from

different datasets. Additionally, enhancements in

hardware capabilities and transmission mechanisms

are necessary to effectively integrate with federated

learning algorithms (Deng, 2019; Deng, 2023;

Sugaya, 2019).

3.2.2 Explainable AI (XAI) for Model

Interpretability

The complexity of DL models often makes them

appear as "black boxes," making clinical adoption

challenging due to the lack of interpretability.

Explainable AI (XAI) aims to make the decision-

making process of these models transparent and

understandable to clinicians. By providing insights

into the models' inner workings, XAI facilitates trust

and enables clinicians to make informed decisions,

ensuring that AI acts as a support tool rather than an

opaque decision-maker.

3.2.3 Bias Mitigation Strategies for

Generalization Across Populations

AI models can inadvertently learn and amplify biases

present in the training data, leading to poor

generalization across different demographics. Bias

mitigation strategies involve techniques for

identifying and reducing these biases during the

model training process. By employing such

strategies, researchers and developers can create AI

systems that perform equitably across diverse patient

populations, ensuring that the benefits of AI in

cardiac diagnosis are accessible to all.

4 CONCLUSIONS

This study details the progress and achievements of

artificial intelligence in the field of cardiac diagnosis,

emphasizing its potential to possess superiority over

traditional methods. This study provides an in-depth

look at various approaches from machine learning to

deep learning models, including their principles,

applications in heart disease diagnosis, and

achievements. These methods include Random Forest

classifier for feature selection, Support Vector

Machine for data classification and other advanced

methods. Also, this thesis lists the challenges and

limitations faced by the field of Artificial Intelligence

in the field of cardiac diagnosis today and also gives

some possible promising solutions.

However, this study can also have some

limitations, for example, ethical considerations and

patient privacy issues related to AI in healthcare can

be further explored. The massive amount of data

required for AI models raises significant privacy

concerns, and this paper would benefit from

discussing how these challenges can be addressed. In

addition, the impact of AI on the healthcare

workforce, including the need for reskilling and

potential job displacement, remains unaddressed but

is a key area for future exploration.

REFERENCES

Butler, L., Ivanov, A., Celik, T., Karabayir, I., Chinthala,

L., Tootooni, M. S., Jaeger, B., Doerr, A., McManus,

D., Davis, R. L., Herrington, D. M., & Akbilgic, O.

2023. Time-Dependent ECG-AI Prediction of Fatal

Coronary Heart Disease. medRxiv.

Deng, X., Oda, S., Kawano, Y., 2023. Graphene-based

midinfrared photodetector with bull’s eye plasmonic

antenna. Optical Engineering, 62(9), p. 097102-

097102.

Deng, X., Li, L., Enomoto, M., Kawano, Y., 2019.

Continuously frequency-tuneable plasmonic structures

for terahertz bio-sensing and spectroscopy. Scientific

reports, 9(1), p.3498.

Dwarakanath, B. L. M. A. R., Jagadish, S., Kallimani,

Ranjan, Walia, & Belete, Birhanu. 2022. A Novel

Feature Selection with Hybrid Deep Learning Based

Heart Disease Detection and Classification in the e-

Healthcare Environment. Computational Intelligence

and Neuroscience.

Guven, M., & Uysal, F. 2023. A New Method for Heart

Disease Detection: Long Short-Term Feature

Extraction from Heart Sound Data. Italian National

Conference on Sensors.

Artiﬁcial Intelligence in Cardiac Disease Diagnosis: A Comprehensive Investigation

139

Hamad, Aqeela, & Jasim, Ammar D. 2021. Heart disease

diagnosis based on deep learning network. Open

Journal of Science and Technology.

Janik, A., Dodd, J., Ifrim, G., Sankaran, K., & Curran, K.

2021. Interpretability of a deep learning model in the

application of cardiac MRI segmentation with an

ACDC challenge dataset. Medical Imaging.

Koch, M., Arlandini, C., Antonopoulos, G., et al. 2023.

HPC+ in the medical field: Overview and current

examples. Technology and Health Care.

Kusuma, S., & JR. 2022. BiDLNet: An Integrated Deep

Learning Model for ECG-based Heart Disease

Diagnosis. International Journal of Advanced

Computer Science and Applications.

Li, S., Kou, P., Ma, M., Yang, H., Huang, S., & Yang, Z.

2024. Application of Semi-supervised Learning in

Image Classification: Research on Fusion of Labeled

and Unlabeled Data. IEEE Access.

Liu, Y., Yang, H., & Wu, C. 2023. Unveiling patterns: A

study on semi-supervised classification of strip surface

defects. IEEE Access, 11, 119933-119946.

Maambo, M., Phiri, J., Kalumbilo, M. M., & Jaganathan, L.

2022. Assisted Artificial Intelligence Medical

Diagnosis System for Heart Disease. Zambia ICT

Journal.

Mohan, S., Thirumalai, C., & Srivastava, G. 2019. Effective

Heart Disease Prediction Using Hybrid Machine

Learning Techniques. IEEE Access, 7, 81542-81554.

https://doi.org/10.1109/ACCESS.2019.2923707

Nedadur, R., Wang, B., & Tsang, W. 2022. Artificial

intelligence for the echocardiographic assessment of

valvular heart disease. Heart.

Prentzas, N., Kakas, A., & Pattichis, C. S. 2023.

Explainable AI applications in the Medical Domain: A

systematic review. arXiv.org.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation

training. Biomedical Signal Processing and

Control, 72, 103323.

Qiu, Y., Chang, C. S., Yan, J. L., Ko, L., & Chang, T. S.

2019. Semantic segmentation of intracranial

hemorrhages in head CT scans. In 2019 IEEE 10th

International Conference on Software Engineering and

Service Science (ICSESS) (pp. 112-115). IEEE.

Saranya, G., & Pravin, A. 2023. Grid Search based

Optimum Feature Selection by Tuning hyperparameters

for Heart Disease Diagnosis in Machine learning. Open

Biomedical Engineering Journal.

Sugaya, T., Deng, X., 2019. Resonant frequency tuning of

terahertz plasmonic structures based on solid

immersion method. 2019 44th International Conference

on Infrared, Millimeter, and Terahertz Waves, p.1-2.

Suresh, T., Assegie, T. A., Rajkumar, S., & Kumar, N. K.

2022. A hybrid approach to medical decision-making:

Diagnosis of heart disease with machine-learning

model. International Journal of Electrical and Computer

Engineering.

Thoenes, M., Agarwal, A., Grundmann, D., Ferrero, C.,

McDonald, A., Bramlage, P., & Steeds, R. 2021.

Narrative review of the role of artificial intelligence to

improve aortic valve disease management. Journal of

Thoracic Disease.

Zhao, F., & Xu, H. 2022. Applications and Current Status

of AI in the Medical Field. Journal of Physics:

Conference Series.

Zhao, F., Yu, F., Trull, T., & Shang, Y. 2023. A new

method using LLMs for keypoints generation in

qualitative data analysis. In 2023 IEEE Conference on

Artificial Intelligence (CAI) (pp. 333-334). IEEE.

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

140