Application of Artificial Intelligence in Stroke Prediction: Latest

Advancements and Future Prospects

Boxiang He

Computer Science, Wenzhou Kean University, Wenzhou, China

Keywords: Artificial Intelligence, Brain Stroke, Machine Learning, Deep Learning.

Abstract: Some scientific and technological methods are more and more applied in medical treatment, and great

achievements have been made in the prediction of stroke, but there are still great challenges. This paper

explores the intersection of stroke management and Artificial Intelligence (AI), focusing on recent

advancements, methodologies, limitations, and future prospects. Stroke, characterized by disrupted blood

flow to the brain, necessitates swift diagnosis and intervention to mitigate potential cell damage or death.

Traditional machine learning algorithms such as Support Vector Machine (SVM) and Random Forest, along

with deep learning algorithms like Artificial Neural Networks (ANN) and Convolutional Neural Networks

(CNN), have been employed to construct predictive models for stroke diagnosis and prognosis. However,

challenges including interpretability, privacy concerns, and applicability across diverse healthcare settings

persist. Solutions such as Shapley Additive Explanations (SHAP), federated learning, and transfer learning

have been proposed to address these challenges and enhance the trustworthiness and generalizability of AI-

driven approaches in stroke management. Continued research efforts are necessary to overcome limitations,

expand sample sizes, and enhance the accuracy and efficiency of AI models in predicting and analyzing

strokes, ultimately improving patient outcomes in stroke management.

1 INTRODUCTION

A stroke, also known as a brain stroke, happens when

blood flow to part of the brain is disrupted, causing

deprivation of oxygen and nutrients, leading to

potential cell damage or death within minutes.

Strokes manifest in various symptoms, such as

weakness, numbness, difficulty speaking, vision

problems, severe headache, and loss of balance.

Recent advancements in deep learning technology

have significantly influenced stroke diagnosis and

treatment in the medical field. Deep learning

algorithms excel in accurately analyzing medical

images to identify stroke lesions, aiding in faster

diagnosis and treatment planning. Additionally, deep

learning can utilize large-scale clinical data to predict

stroke occurrence, enabling early intervention and

prevention (Lee, 2017). Pre-trained models further

enhance medical image analysis, improving

diagnostic accuracy and efficiency. In summary, the

integration of deep learning technology offers

promising prospects for precise and effective stroke

https://orcid.org/0009-0001-9327-5761

diagnosis, treatment, and prevention, potentially

improving patient outcomes.

Recent advancements in the intersection of brain

strokes and Artificial Intelligence (AI) showcase a

promising frontier in healthcare (Qiu, 2022). AI

technologies are being increasingly employed across

various facets of stroke management, catalyzing

significant improvements in patient care. Initially, AI

algorithms are revolutionizing stroke diagnosis by

swiftly and accurately analyzing medical imaging

scans, including Magnetic Resonance Imaging

(MRIs) and Computed Tomography (CT) scans. This

capability enables clinicians to promptly identify

stroke symptoms and initiate timely interventions,

thereby potentially minimizing long-term damage.

Moreover, machine learning models harness vast

datasets of patient records to predict individual stroke

risks, facilitating proactive measures for prevention

and intervention (Soun, 2021). Additionally, AI

systems leverage real-time patient data and medical

literature to generate personalized treatment plans

tailored to each patient's unique needs. By optimizing

He, B.

Application of Artiﬁcial Intelligence in Stroke Prediction: Latest Advancements and Future Prospects.

DOI: 10.5220/0012960700004508

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 617-621

ISBN: 978-989-758-713-9

617

treatment strategies, AI contributes to enhancing

patient outcomes and reducing healthcare costs (Haris,

2018). According to a 2017 study, machine learning

has been widely used in stroke imaging, including the

two main methods of Artificial Neural Network

(ANN) and Convolutional Neural Network (CNN). In

addition, according to a 2018 study, artificial

intelligence is also used to analyze data on various

brain diseases. great effect, but still has limitations

Furthermore, AI-driven rehabilitation tools are aiding

stroke survivors in regaining motor function and

improving overall recovery trajectories. These tools,

which incorporate robotics, virtual reality, and

personalized exercise regimens, offer tailored support

to each patient's rehabilitation journey. Lastly, AI-

enabled telemedicine platforms are facilitating

remote monitoring of stroke patients, allowing

healthcare providers to deliver timely interventions

and support, even in underserved areas. This remote

monitoring capability minimizes the need for in-

person visits, thereby increasing accessibility to

quality stroke care. In summary, the integration of AI

into stroke management holds immense promise for

revolutionizing diagnosis, treatment, rehabilitation,

and remote monitoring, ultimately leading to better

patient outcomes and healthcare delivery (Dritsas,

2022).

The remainder of this article will delve into the

method, discussion, results, and conclusion. Firstly,

the method section will meticulously review and

encapsulate the research methodologies employed in

studying the intersection of stroke and AI in recent

years. The forthcoming investigation will undertake a

meticulous examination of diverse methodologies

employed within the realms of stroke detection,

prediction, treatment optimization, rehabilitation, and

telemedicine. This scrutiny will encompass an

analysis of the datasets utilized, the algorithms

deployed, and the evaluation criteria employed across

these studies. Subsequently, the ensuing discussion

segment will undertake a critical evaluation of these

methodologies, delineating their respective strengths

and weaknesses. Particular emphasis will be placed

on elucidating the encountered limitations,

identifying potential areas for enhancement, and

envisioning future prospects for furthering

technological advancements. This discourse will

encompass a comprehensive exploration of the

challenges inherent in the integration of artificial

intelligence into stroke management practices, whilst

offering insights into strategies to mitigate these

challenges and augment the efficacy of AI-driven

approaches. Lastly, the concluding remarks will

consolidate and synthesize the key findings and

insights derived from the preceding sections. This

will involve furnishing a comprehensive overview of

the paper's contributions to the academic domain,

delineating avenues for prospective research

endeavors, and underscoring the imperative nature of

ongoing innovation in harnessing AI to ameliorate

stroke care practices.

2 METHOD

2.1 Traditional Machine Learning

Algorithms

In recent years, machine learning has rapidly

developed and developed in a variety of applications

in various healthcare systems (Bi, 2019). Machine

learning is a branch of artificial intelligence that

focuses on giving computer systems the ability to

learn and improve automatically without explicit

programming instructions. It can extract patterns and

regularities from large amounts of data to

autonomously infer, generalize, and predict future

behavior. Its goal is to improve task performance by

letting machines learn from experience. Machine

learning is of great help in the treatment and research

of stroke. It helps analyze data and build models to

facilitate doctors to treat faster and more accurately.

Early detection of stroke is a critical step in effective

treatment, and machine learning can be of huge value

in this process, which is the ultimate technology that

can help medical professionals make clinical

decisions and predictions (Sirsat, 2020).

2.1.1 Random Forest

Random forest is an integrated learning method that

integrates the prediction results of each tree to

improve the accuracy and robustness of the model by

constructing multiple decision trees and using

random sampling and random feature selection. It is

suitable for classification and regression tasks. And it

has the characteristics of high parallelism and

resistance to overfitting. The application of random

forest in stroke research mainly includes two aspects:

prediction and diagnosis of stroke, and rehabilitation

and prognosis assessment after stroke. In terms of

prediction and diagnosis, random forests can use

patients' clinical data, imaging characteristics, etc. to

predict stroke risk and diagnose conditions,

improving the accuracy and timeliness of diagnosis

(Steven, 2017). In addition, it can also be used to

guide the formulation and evaluation of rehabilitation

treatment plans, thereby improving the rehabilitation

effect and prognosis quality. These applications

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

618

provide important support and guidance for early

intervention, personalized treatment and

rehabilitation of stroke (Carlos, 2021).

2.1.2 Support Vector Machine (SVM)

Support Vector Machine (SVM) is a machine learning

method commonly used for classification and

regression analysis. The main idea is to find an

optimal hyperplane that separates sample points of

different categories.

In classification problems, the goal of SVM is to

find a decision boundary that can move samples of

different categories as far away from the hyperplane

as possible, which can increase the robustness of

classification. In order to achieve this goal, SVM

defines the decision boundary through support

vectors (support samples), which are the sample

points closest to the hyperplane. SVM determines the

optimal decision boundary by maximizing the

distance (i.e., margin) between the support vector and

the hyperplane. This distance is also called margin

maximization. In a 2017 study, there were examples

of SVM applied to brain imaging, which is also a

possibility that SVM can be applied to stroke

(Cuingnet, 2010).

2.2 Deep Learning Algorithms

Moreover, in recent years, deep learning algorithms

have also been of great help in the study of stroke.

Deep learning algorithm is an artificial intelligence

technology that learns the characteristics and patterns

of data through a multi-level neural network structure.

These neural networks are composed of a large

number of neurons, each of which is connected to the

neurons of the previous layer. Through an iterative

process, the connection weights between neurons are

constantly adjusted to maximize the accuracy of

predicting or classifying the input data. Through deep

learning algorithms, the system can automatically

discover complex patterns in data and perform

efficient feature extraction and abstraction to achieve

various intelligent tasks such as image recognition,

speech recognition, and natural language processing.

Wang et al. have applied deep learning algorithms in

pathological image analysis in 2019 (Wang, 2019).

These algorithms not only have high accuracy, but

also have high computational efficiency. These

methods can also be applied to analyze images of

strokes.

2.2.1 ANN

In the treatment and research of stroke, ANN has also

played a certain role. It represents a computing model

that imitates the structure and function of biological

neural networks. Artificial neural networks are

composed of a large number of artificial neurons (or

nodes) that are connected to each other through

connections (or edges) to form a network. Each

neuron receives input signals from other neurons,

performs a weighted summation of these signals by

weights, and then passes the result to an activation

function for processing, ultimately producing an

output. ANN is usually trained through optimization

methods such as backpropagation algorithm and

gradient descent to adjust the connection weights

between neurons so that the network can learn and

adapt to the patterns and characteristics of the input

data to achieve various tasks such as classification,

regression, Clustering etc. In 2019, Chen et al.

analyzed the feasibility of ANN for stroke risk

stratification and concluded that ANN is very

effective in predicting stroke (Chan, 2019).

The structure of an Artificial Neural Network

consists of the input layer, hidden layers, and output

layer. The input layer receives external input data,

with each node representing a feature of the input

data. The hidden layers, composed of multiple layers

of neurons, learn the complex relationships between

input data through adjustment of connection weights

and bias parameters. Eventually, the output layer

generates the final output of the model, with the

number of nodes determined by the type of task the

model performs. During training, ANN utilizes the

backpropagation algorithm to continuously adjust

connection weights and biases to minimize the loss

function and enhance the model's performance. This

network structure and learning algorithm enable

ANN to effectively handle various types of data,

performing prediction and classification tasks, and

find wide applications across multiple domains,

including stroke analysis.

2.2.2 CNN

Convolutional Neural Network is an artificial neural

network model that specializes in processing grid-

structured data (such as images, audio). Its core

components include convolutional layers, pooling

layers and fully connected layers. The convolutional

layer extracts the features of the input data by sliding

the convolution kernel, the pooling layer reduces the

size of the feature map and enhances the translation

invariance of the model, and finally performs

classification or other tasks through the fully

connected layer. CNN has achieved great success in

the field of computer vision and is widely used in

tasks such as image recognition and target detection.

Application of Artiﬁcial Intelligence in Stroke Prediction: Latest Advancements and Future Prospects

619

It has also demonstrated promise in areas like natural

language processing, particularly in the analysis of

images related to strokes. Identifying the location of

an ischemic stroke in CT images can be challenging,

as it's not always clearly visible. Consequently,

diagnosis often depends on a physician's assessment

of the images. CNN can be highly effective in aiding

with this task. In 2017, Lin et al. studied the accuracy

of CNN in identifying ischemic stroke. The accuracy

rate was as high as 90%, which shows that CNN has

great potential in predicting cerebral stroke (Chin,

2017).

The structure of a Convolutional Neural Network

comprises several key components. Firstly,

convolutional layers extract features from input data

using filters, generating feature maps. Pooling layers

reduce spatial dimensions while preserving essential

information. Fully connected layers connect extracted

features to the output layer for classification or

regression tasks. Activation functions introduce non-

linearity to the network. Batch normalization layers

enhance training stability by normalizing feature

maps. Through the combination and stacking of these

components, CNNs effectively extract features from

input data, enabling efficient processing and learning

of complex data such as images and audio.

3 DISCUSSIONS

Although significant progresses have been made, the

application of AI in the context of brain stroke

diagnosis and treatment presents several limitations

and challenges. Firstly, interpretability of AI models

remains a significant concern, particularly in complex

medical scenarios such as stroke diagnosis, where

clinicians require clear explanations for model

predictions to make informed decisions. Secondly,

privacy concerns arise due to the sensitive nature of

medical data, including patient imaging scans and

health records, necessitating robust data protection

measures. Thirdly, the applicability of AI algorithms

across diverse healthcare settings and patient

populations poses challenges due to variations in data

quality, accessibility, and clinical practices. Fourth,

there are limitations in the research on predicting risk

factors for various types of strokes (Bandi, 2020).

To address these challenges, several solutions

have been proposed. For interpretability, techniques

such as Shapley Additive Explanations (SHAP)

provide insights into model predictions, enabling

clinicians to understand the rationale behind AI

recommendations. Privacy concerns can be mitigated

through the implementation of privacy-preserving AI

approaches, including techniques such as federated

learning, which allows model training on

decentralized data without exchanging raw data

between institutions. Additionally, ensuring the

applicability of AI algorithms involves developing

adaptable models that can accommodate variations in

data sources and clinical contexts, as well as fostering

collaborations between AI researchers and healthcare

professionals to tailor solutions to specific clinical

needs. In addition, some models such as random

forests can also help to accurately predict risk factors,

and it is possible to use image datasets to derive

different types of strokes and risk levels in the future

(Bandi, 2020).

Looking ahead, the prospects of AI in brain stroke

management are promising. SHAP and similar

interpretability techniques will continue to evolve,

providing deeper insights into AI decision-making

processes and enhancing trust in AI-driven clinical

decision support systems. Expert systems that

combine domain expertise with AI algorithms hold

potential for personalized stroke management by

integrating clinical guidelines and patient-specific

data. Federated learning offers a pathway for

collaborative model development across institutions

while preserving data privacy, facilitating the creation

of robust and generalizable stroke prediction models.

Furthermore, transfer learning techniques enable the

transfer of knowledge from related tasks or domains

to improve the performance of AI models in stroke

diagnosis and prognosis. Overall, the continued

advancement and integration of these AI approaches

hold great promise for enhancing stroke care

outcomes and reducing the burden of this devastating

neurological condition.

4 CONCLUSIONS

In this work, this paper summarized research on AI's

prediction of stroke and how it can help in this regard.

In the previous prediction and analysis of stroke,

various traditional machine learning (SVM and

random forest) and deep learning algorithms (ANN

and CNN) were used to construct a model that

improved the accuracy and efficiency of prediction.

In the process, it can be found that there are still some

limitations and deficiencies, and the sample size is not

comprehensive, which means that in the future, the AI

model for predicting and analyzing stroke can be

improved, and more samples can be obtained to

improve the accuracy of prediction.

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

620

REFERENCES

Bandi, V., Bhattacharyya, D., & Midhunchakkravarthy, D.

2020. Prediction of Brain Stroke Severity Using

Machine Learning. Rev. d'Intelligence Artif., 34(6),

753-761.

Bi, Q., Goodman, K. E., Kaminsky, J., & Lessler, J. 2019.

What is machine learning? A primer for the

epidemiologist. American journal of epidemiology,

188(12), 2222-2239.

Chan, K. L., Leng, X., Zhang, W., Dong, W., Qiu, Q., Yang,

J., ... & Liu, J. 2019. Early identification of high-risk

TIA or minor stroke using artificial neural network.

Frontiers in Neurology, 10, 171.

Chin, C. L., Lin, B. J., Wu, G. R., Weng, T. C., Yang, C. S.,

Su, R. C., & Pan, Y. J. 2017. An automated early

ischemic stroke detection system using CNN deep

learning algorithm. In 2017 IEEE 8th International

conference on awareness science and technology

(iCAST) (pp. 368-372). IEEE.

Cuingnet, R., Rosso, C., Lehéricy, S., Dormont, D., Benali,

H., Samson, Y., & Colliot, O. 2010. Spatially

regularized SVM for the detection of brain areas

associated with stroke outcome. In Medical Image

Computing and Computer-Assisted Intervention–

MICCAI 2010: 13th International Conference, Beijing,

China, September 20-24, 2010, Proceedings, Part I 13

(pp. 316-323). Springer Berlin Heidelberg.

Dritsas, E., & Trigka, M. 2022. Stroke risk prediction with

machine learning techniques. Sensors, 22(13), 4670.

Fernandez-Lozano, C., Hervella, P., Mato-Abad, V.,

Rodríguez-Yáñez, M., Suárez-Garaboa, S., López-

Dequidt, I., ... & Iglesias-Rey, R. 2021. Random forest-

based prediction of stroke outcome. Scientific reports,

11(1), 10071.

Kamal, H., Lopez, V., & Sheth, S. A. 2018. Machine

learning in acute ischemic stroke neuroimaging.

Frontiers in neurology, 9, 410345.

Lee, E. J., Kim, Y. H., Kim, N., & Kang, D. W. 2017. Deep

into the Brain: Artificial Intelligence in Stroke Imaging.

Journal of stroke, 19(3), 277–285.

https://doi.org/10.5853/jos.2017.02054

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.

Rigatti, S. J. 2017. Random forest. Journal of Insurance

Medicine, 47(1), 31-39.

Sirsat, M. S., Fermé, E., & Câmara, J. 2020. Machine

learning for brain stroke: a review. Journal of Stroke

and Cerebrovascular Diseases, 29(10), 105162.

Soun, J. E., Chow, D. S., Nagamine, M., Takhtawala, R. S.,

Filippi, C. G., Yu, W., & Chang, P. D. 2021. Artificial

intelligence and acute stroke imaging. American

Journal of Neuroradiology, 42(1), 2-11.

Wang, S., Yang, D. M., Rong, R., Zhan, X., & Xiao, G.

2019. Pathology image analysis using segmentation

deep learning algorithms. The American journal of

pathology, 189(9), 1686-1698.

Application of Artiﬁcial Intelligence in Stroke Prediction: Latest Advancements and Future Prospects

621