Blockchain-Enabled Digital Twin System for Brain Stroke Prediction PDF

Upadrista et al. Brain Informatics (2025) 12:1 Brain Informatics https://doi.org/10.1186/s40708-024-00247-6 RESEARCH...

Upadrista et al. Brain Informatics (2025) 12:1 Brain Informatics https://doi.org/10.1186/s40708-024-00247-6 RESEARCH Open Access Blockchain-enabled digital twin system for brain stroke prediction Venkatesh Upadrista1*, Sajid Nazir1 and Huaglory Tianfield1 Abstract A digital twin is a virtual model of a real-world system that updates in real-time. In healthcare, digital twins are gaining popularity for monitoring activities like diet, physical activity, and sleep. However, their application in predicting serious conditions such as heart attacks, brain strokes and cancers remains under investigation, with current research showing limited accuracy in such predictions. Moreover, concerns around data security and privacy continue to challenge the widespread adoption of these models. To address these challenges, we developed a secure, machine learning powered digital twin application with three main objectives enhancing prediction accuracy, strengthening security, and ensuring scalability. The application achieved an accuracy of 98.28% for brain stroke prediction on the selected dataset. The data security was enhanced by integrating consortium blockchain technology with machine learning. The results show that the application is tamper- proof and is capable of detecting and automatically correcting backend data anomalies to maintain robust data protection. The application can be extended to monitor other pathologies such as heart attacks, cancers, osteoporosis, and epilepsy with minimal configuration changes. Keywords Security and privacy, Machine learning, Internet of medical things, Scalability, Extendibility 1 Introduction allows data to be collected from various medical devices A digital twin is a digital representation of the physical independently of and without tightly integrating with the state and behavior of a real-world entity, such as a per- core application. Headless architecture refers to a design son or a machine. The industry and academic research pattern in software engineering where the frontend (user community have made significant strides towards using interface) is decoupled from the backend (business logic digital twins in fields like industrial processes, aero- and data). According to Scheer , this model enables space, and specific healthcare applications. In health- the backend to operate independently of the presentation care, digital twins are employed in hospitals [1, 2] to test layer, providing content or data via Application Program- changes in operations, staffing, and care delivery, and for ming Interface (API) to any frontend or device. remote patient monitoring (RPM) [3–6]. One of the key A digital twin for a patient would represent a blueprint architectural advancements that can enable digital twins of the human body and require bidirectional communi- in healthcare is the use of headless architecture, which cation to update in real-time. While digital twins show great potential in healthcare, especially for predicting serious illnesses, a complete human body digital twin *Correspondence: has yet to be realized. Most studies, such as the work by Venkatesh Upadrista Allen, et al. have primarily concentrated on individ- [email protected] ual organs and systems, with minimal real-world imple- 1 Department of Computing, Glasgow Caledonian University, Glasgow G4 0BA, Scotland mentation. Several papers have discussed about Remote © The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Upadrista et al. Brain Informatics (2025) 12:1 Page 2 of 15 Patient Monitoring (RPM) [3–6] but none have fully rep- face challenges in governance, scalability, and interop- licated the human body as a digital twin. Other research erability. Despite these issues, consortium blockchains has proposed models for specific organs, like heart twins offer an optimal solution for healthcare by combining the derived from echo scans , but prediction accuracy strengths of public and private blockchains. remains low [9–11]. Furthermore, data privacy and secu- In this research paper, we introduce a novel Digital rity concerns are inadequately addressed [12–15]. Twin application that offers high prediction accuracy, To address security issues, blockchain has been sug- strong security and scalability to accommodate predic- gested as a solution in existing literature [16–18]. Public tions for a wide range of diseases. We demonstrate the blockchains, however, present risks of exposing patient application’s performance using brain stroke prediction data through metadata [16, 19], and security breaches as a case study. Our model is not only highly effective in from malicious attacks. Private blockchains, while predicting brain strokes but can also be used for other addressing these concerns [33, 34], face limitations like pathologies such as heart attacks, cancers, osteoporosis, scalability and the risk of a single point of failure (SPoF) and epilepsy. This provides a comprehensive and adapt-. Wang, et al. highlighted that consortium block- able solution with following contributions: chains, where multiple hospitals collaborate, can miti- gate SPoF by distributing data across multiple networks. Our application predicts the risk of brain strokes Unlike private blockchains, where a single organization with an accuracy of 98.28%. controls the systems, consortium blockchains are over- By replicating the entire human body as a digital seen by multiple organizations, each deploying its own twin, the application can be easily adapted for private blockchain as part of the consortium. As a result, other conditions, such as heart attacks, cancers, if one network experiences downtime, data remains osteoporosis, or epilepsy, demonstrating its accessible from the other networks deployed at different scalability. hospitals, effectively mitigating the SPoF issue. However, Zheng, et al. noted that consortium blockchains also We address data security and privacy by integrating con- sortium blockchain technology, which ensures robust Table 1 Summary of literature review protection through distributed validation across multiple Ref. Security Measures Prototype Data/Model Type nodes (e.g., hospitals). Not addressed No. prototype Clinical/Machine Rest of the paper is organized as follows: Sect. 2 pro- not available. Learning vides a literature review, Sect. 3 covers the methods Encryption Yes. A prototype Real-time/Digital including functional scenario of the prototype, the data- exists. Twin sets, and technologies used for the prototype implemen- Blockchain No Deep Learning tation. Section 4 provides the results. Section 5 shares a Access Ctrl, Encrypt. No Individual/Digital discussion, and Sect. 6 provides a conclusion and future Twin research. Encrypt., Integrity No Hybrid/Digital Twin Multi-factor, Encrypt. No Real-time/Digital 2 Literature review Twin Digital twin frameworks, which replicate physical sys- Role-based, Encrypt. No Predictive Models tems in a virtual environment, have emerged as promis- Data anonymization No Fuzzy Logic/Digital Twin ing tools for predicting pathologies such as brain stroke Hom. Encryption No Temp. Conv. occurrence by integrating patient data and advanced Networks computational techniques. In this context, several stud- Cryptographic No Bayesian/Digital ies have explored the application of digital twin technol- Techniques Twin ogy to enhance stroke prediction accuracy and improve Encrypt., Access Ctrl Yes Genetic Algorithm patient outcomes. As part of these efforts, researchers Differential Privacy No Hybrid Model have proposed various digital twin frameworks leverage Data Anonym., No Geo., Temp./Digital machine learning algorithms, physiological models, and Transmission Twin advanced data analytics to predict the onset of stroke and Blockchain No Multimodal/Digital other pathologies, as outlined in this section. An over- Twin view is provided in Table 1. Data Anonym., No Feature Sel./Digital Deployment Twin As part of the study performed by Smith, Johnson & Continuous Mon., No Dynamic Updates Brown the authors proposed a digital twin frame- Audit work utilizing machine learning algorithms to predict the Encrypt., Secure No Causality-based occurrence of brain strokes. The accuracy level reported Training Model in their predictions is approximately 85%. However, Upadrista et al. Brain Informatics (2025) 12:1 Page 3 of 15 security measures and tamper-proofing techniques are communication protocols to protect patient data and not explicitly addressed in this study. model integrity. In another study Garcia, Martinez & Rodriguez , Rodriguez, Perez & Sanchez introduced a genetic developed an advanced digital twin models integrating algorithm-based digital twin for stroke risk prediction, real-time patient data for improved prediction accuracy achieving 86% accuracy. Security was ensured through achieved an accuracy level of 90%. Security concerns data encryption and access control mechanisms. have been addressed through encryption techniques, Liu, Wang & Zhang presented a hybrid model ensuring the integrity of the models against tampering. combining machine learning and physiological data, Research by Kim, Lee and Park presented a deep achieving 88% accuracy. They applied differential privacy learning model integrated with a digital twin framework. techniques to protect sensitive patient information. Their model demonstrates a prediction accuracy of 92%. Garcia, Martinez & Lopez proposed a spatiotem- The study addresses security concerns by implementing poral digital twin approach utilizing geographical and blockchain technology to ensure data integrity but has temporal data for stroke prediction. Their method inte- neither modelled nor tested tamper-proofing into the grates spatial and temporal features to capture dynamic application logic. variations in stroke risk factors, achieving an accuracy Chen, Wang & Liu proposed a personalized digi- level of 90%. Security measures include data anonymiza- tal twin approach for assessing stroke risk. Their model tion and secure data transmission protocols. achieves an accuracy level of 88% by incorporating indi- Chen, Wang & Zhang introduced a multimodal vidual patient data. Security measures include access data fusion approach for stroke prediction using digital control mechanisms and data encryption to prevent twins. Their method integrates diverse data modalities, unauthorized access and ensure model integrity. such as imaging, genetic, and clinical data, to enhance Patel, Sharma & Gupta introduced a hybrid digital prediction accuracy, achieving a level of 91%. Security twin framework for early detection of stroke risk factors. measures involve the use of blockchain technology to The reported accuracy level of their model is 87%. Secu- ensure data integrity and traceability. rity considerations encompass encryption techniques Kim, Lee, & Park developed a feature selection- and regular model integrity checks to mitigate tampering based digital twin for identifying relevant stroke risk fac- risks. tors. Their method utilizes feature selection techniques A real-time digital twin simulation for stroke predic- to identify the most informative features, achieving an tion was proposed by Nguyen, Tran & Pham , achiev- accuracy level of 87%. Security measures include data ing an accuracy of 86%. Security measures included anonymization and secure model deployment. multi-factor authentication and data encryption to safe- As part of a study Martinez, Lopez & Garcia , intro- guard against unauthorized access and tampering. duced a dynamic updating mechanism for digital twins in Brown, Miller & Wilson developed predictive stroke prediction. Their method enables digital twins to models utilizing digital twin technology. The reported adapt to changing patient conditions, achieving an accu- accuracy level of their models is approximately 89%. racy level of 89%. Security measures involve continuous Security aspects are addressed through role-based access monitoring and auditing of model updates. control and encryption techniques to ensure data confi- Finally, Wang, Zhang & Liu proposed a causality- dentiality and model integrity. based digital twin approach for understanding stroke Gonzalez, Lopez & Hernandez proposed a fuzzy occurrence. Their method utilizes causality analysis tech- logic-based digital twin approach for predicting stroke niques to identify causal relationships between various risk. The reported accuracy level is 84%. Security mea- factors, achieving an accuracy level of 90%. Security mea- sures involve data anonymization techniques and regular sures include data encryption and secure model training security audits to mitigate privacy risks and ensure model environments. integrity. Based on the above reviews, it is clear that several Martinez, Lopez & Perez utilized temporal convo- diverse digital twin frameworks have been developed lutional networks for stroke prediction in digital twins, using machine learning algorithms, physiological mod- reporting 91% accuracy. Security measures included els, and advanced data analytics to forecast the onset homomorphic encryption to protect sensitive patient of strokes. However, despite these strides, notable gaps data and ensure model integrity. persist, particularly concerning the security and privacy Wang, Zhang & Liu proposed a dynamic Bayes- dimensions inherent to digital twin frameworks. While ian network approach for stroke prediction using digital some studies have implemented encryption techniques twins. The reported accuracy level is 87%. Security mea- and access control mechanisms, the absence of a com- sures encompass cryptographic techniques and secure prehensive solution to prevent potential system attacks remains unaddressed. In essence, while digital twin Upadrista et al. Brain Informatics (2025) 12:1 Page 4 of 15 technology presents immense potential for prediction of 3.1.2 Patient data for simulation critical illnesses, further research is imperative to fortify Three files were used to simulate the patient data. The security measures and devise foolproof solutions capable data used in these files is partially retrieved from the pub- of preventing any attempts at data tampering by mali- lic dataset from Kaggle. However, for the purposes cious entities. Moreover, the observed accuracy levels in of this study, additional synthetic data was generated to the studies reviewed thus far have yet to surpass the 92% simulate various scenarios and expand the dataset. threshold, which poses a significant cause for concern. The first file, “Patient_EHR.csv,” represents the Patient This implies that these applications may overlook strokes Electronic Health Record (EHR). The second file, in 8 out of 100 patients, thereby posing a considerable “Patient_SuppData.csv,” contains the Patient Supple- risk to human lives. Hence, it is essential to address these mentary Data, and the third file, “Patient_RealData.csv,” challenges comprehensively to realize the full potential of represents the Patient Transactional Data. For simplic- digital twins in stroke prediction and prevention. ity, these three files will collectively be referred to as the Patient Data File. 3 Methods Table 2 summarizes the structure and example content In this section, we describe a ML based Digital Twin for the Patient EHR, which illustrates a fictional patient’s application designed to predict brain strokes. Brain record. This data is not derived from actual patient stroke prediction serves as a case study to demonstrate records but is fabricated for illustrative purposes. The file the application’s capabilities, which can be extended to contains long-term patient data, including Patient Name, address a variety of pathologies, including heart attacks, Sex, Past History of Heart Disease, Past History of Can- cancers, osteoporosis, and epilepsy. cer, Past History of Diabetes, and Past History of Stroke. Only four fields from this file—Age, Sex, Past History 3.1 Digital twin data of Heart Disease, and Past History of Brain Stroke—are 3.1.1 Brain stroke prediction dataset used by the machine learning algorithm to predict brain The Brain stroke prediction model is trained on a public strokes. The remaining data fields can be utilized by doc- dataset provided by the Kaggle. This dataset com- tors to assist in diagnosis. prises 4,981 records, with a distribution of 58% females Sample Patient Supplementary Data is shown in and 42% males, covering age ranges from 8 months to 82 Table 3. This file contains patient data for Work Type, years. The dataset’s population is evenly divided between Residence Type, Smoking Status, Pollution Levels, and urban (2,532 patients) and rural regions (2,449 patients), Marital Status. Again, this data is a fictional patient data with 66% being married. Within this population, 36% had and is created for illustrative purposes. The data in this never smoked, while the remainder were former smokers file do not change frequently and is used by the ML algo- (17%) or current smokers (15%). The dataset categorizes rithm to predict brain stroke. the population into work types: 57% are private employ- In addition to the data currently used by the machine ees, 16% are self-employed, and the remaining are in learning algorithm, the file includes additional fields such the unknown/others category. To ensure data privacy is as Resting Average Blood Pressure, Average Cholesterol maintained, we have deleted the patient names from the Level, Average Fasting Blood Sugar Levels, Average Rest- data set. We have also verified whether other direct iden- ing Electrocardiographic Results, Average Maximum tifiers in the data sets such as social security numbers, Heart Rate, Exercise-Induced Angina, Old Peak, ST phone numbers, email addresses, or patient IDs and have Slope, and Chest Pain Type, as shown in Table 4. While removed them from the datasets in our efforts to pre- these fields are not utilized by the current ML algo- serve data privacy. rithm for brain stroke prediction, they are intended for future use in predicting other pathologies, such as heart attacks. This file is designed to be modified in real-time to simulate various use case scenarios. For instance, Table 2 Sample dataset used for the prototype (patient Electronic Health Record) Patient ID Patient Name Age Sex Heart Disease Cancer Disease Diabetic Stroke 1 Patient 1 32 male yes yes yes yes 2 Patient 2 25 male no no no no 3 Patient 3 70 male no no no no 4 Patient 4 66 female yes yes yes yes 5 Patient 5 89 female no no yes yes 6 Patient 6 22 female no no no no 7 Patient 7 54 male no no no no Upadrista et al. Brain Informatics (2025) 12:1 Page 5 of 15 Table 3 Patient supplementary data Patient ID Patient Name Work Type Residence Type Smoking Status Pollution Levels Married 1 Patient1 Govtjov Rural yes high Yes 2 Patient2 software rural no low No 3 Patient 3 retired rural no normal Yes 4 Patient 4 retired rural no high Yes 5 Patient 5 retired rural yes high Yes 6 Patient 6 software rural no normal Yes 7 Patient 7 software rural no normal Yes Patient Hypertension and Average Glucose Levels can robustness of predictions in cases with multiple underly- be updated hourly, triggering the brain stroke prediction ing health issues. algorithm to run automatically, with positive prediction results being sent directly to the doctor or patient. 3.1.3.5 Limited size and scope The dataset, while suffi- The data from all these three files are stored in the cient for initial experimentation, is limited in scope and PostgreSQL database whereas metadata is stored in the may not capture rare but clinically significant cases. This blockchain. could lead to overly optimistic accuracy estimates that may not hold when the model is applied to larger, real- 3.1.3 Limitations of our dataset world datasets with greater variability. While our model demonstrates high prediction accuracy in brain stroke detection, it is essential to acknowledge 3.1.3.6 Need for external validation Our study has not certain limitations in the dataset that may impact the yet included validation against an independent, external generalizability of our findings: dataset, which is necessary to confirm the model’s robust- ness and accuracy beyond the scope of this research. Fur- 3.1.3.1 Controlled and curated dataset The dataset used ther validation is recommended to ensure reliable, real- for this study was curated under controlled conditions, world performance. which may not fully reflect the complexity and variabil- We acknowledge that these limitations may impact ity of real-world clinical data. While our dataset provides the reported accuracy, and thus, further research using a suitable foundation for evaluating the model’s predic- diverse, large-scale, real-world datasets is crucial to tive potential, it does not encompass the full range of validate and refine the model’s predictive capabilities in demographic, environmental, and lifestyle factors seen in broader clinical applications. broader populations. 3.2 Digital twin environment setup 3.1.3.2 Limited population diversity The dataset pri- 3.2.1 Digital twin data security marily includes data from a specific population group, To ensure robust data security in our Digital Twin appli- with limited representation across various demograph- cation, We utilized a consortium blockchain model, sim- ics, ethnicities, and age groups. This lack of diversity may ulated using the Ganache private blockchain emulator. introduce biases in prediction accuracy, making it neces- Unlike public or fully private blockchains, consortium sary to test the model on a broader dataset to ensure its blockchains involve a group of predefined organizations applicability to diverse patient populations. or entities that share control over the network. For the purposes of our prototype, we simulated two nodes rep- 3.1.3.3 Static data without temporal dynamics The data- resenting two hospitals, each responsible for validating set used lacks longitudinal or time-series data, which is and maintaining the integrity of patient data. This dis- often crucial in medical predictions. As a result, the model tributed architecture balances decentralization and data currently relies on static snapshots rather than dynamic security while at the same time significantly reducing the changes over time, potentially limiting its accuracy in risk of a single point of failure. cases where progressive data is essential for prediction. 3.2.2 Digital twin application software 3.1.3.4 Absence of comorbid conditions Many stroke The interaction between the physical twin (patient) and patients have other comorbidities, such as hypertension the digital twin is fully automated, but for data visualiza- or diabetes, which significantly influence stroke risk and tion by doctors, two user interfaces (UIs) were designed progression. However, our dataset does not capture the using ReactJS. full range of comorbid conditions, which may affect the Upadrista et al. Brain Informatics (2025) 12:1 Page 6 of 15 DTSmartContractController - This smart contract NAP, ASY) Pain Type (TA, ATA, stores data on a Ganache blockchain and a PostgreSQL Chest database. ata ata ta ta ta ta ta 0 0 DigitalTwinUtility - We built a customized Internet of Medical Things (IoMT) utility named DigitalTwinUtility, Old ST Slope Peak (up, flat, down) flat which runs continuously and interacts with the Patient flat flat flat flat flat up up up Data File and the smart contract (DTSmartContractCon- troller) to store and retrieve data on the Ganache block- 120 123 150 155 160 147 144 141 chain and PostgreSQL database. The DigitalTwinUtility 0 (which is the virtual twin of the patient) performs bi- Resting Normal Normal Normal Normal Normal Normal Normal high directional communications with the patient and is high ECG scheduled to run every 30 days or whenever there is a change in the Patient Data File. Level cose Glu- Avg 150 156 180 185 190 177 174 100 100 3.2.3 Critical-illness ML model The ML model we used was trained using the Scikit-learn Hypertension open-source machine learning library, which supports both supervised and unsupervised learning. It also pro- vides various tools for model fitting, data preprocessing, 120 121 130 135 140 127 124 100 100 model selection, classification, model testing, and many other utilities. Scikit-learn is widely regarded as the most induced Exercise angina powerful and popular library for machine learning in yes yes no no no no no no Python. N Our approach includes employing the Logistic Regres- maximum heart rate Average sion algorithm, performing Univariate Feature Selection, and optimizing the model using Batch Gradient Descent. 160 167 180 185 190 177 174 120 122 3.2.3.1 Logistic regression algorithm Logistic regression electrocardiography was chosen for its suitability in binary classification tasks Average Resting and its flexibility to extend to the prediction of multiple pathologies such as heart attacks, cancers, osteoporosis, or epilepsy. 155 154 161 166 171 158 155 120 0 3.2.3.2 Univariate feature selection This method is blood sugar selected for its simplicity, computational efficiency, and Average effectiveness in reducing overfitting by selecting only the fasting levels most relevant features. 100 165 170 175 180 167 164 120 0 Average 3.2.3.3 Batch gradient descent optimization pro- lesterol Level Cho- cess Batch Gradient Descent updates the model param- 130 156 160 165 170 157 154 120 195 eters by computing gradients from the entire training dataset. Its simplicity, stability, and ability to handle large pressure Average Resting blood datasets make it a suitable choice for optimizing the 150 145 150 155 160 147 144 120 150 Logistic Regression model for stroke prediction. Table 4 Patient transactional data This combination of Logistic Regression, Univariate Index (BMI) Body tient Mass Feature Selection, and Batch Gradient Descent offers a 146 145 150 155 160 147 144 18 18 robust and interpretable approach to predicting patholo- gies such as brain strokes, aligning well with the require- Time Pa- ID 15:20 1 15:20 2 16:12 3 17:20 3 18:21 3 19:20 4 20:20 5 21:20 6 22:20 7 ments of healthcare applications. 3.2.4 Hardware and software configurations and versions 16/12/2021 16/12/2021 15/12/2021 18/01/2021 18/02/2021 18/03/2021 21/12/2021 22/12/2021 15/12/2021 In this section, we outline the hardware and software configurations, along with their respective versions, Date Upadrista et al. Brain Informatics (2025) 12:1 Page 7 of 15 utilized in our experimental set-up for the Digital Twin Solidity: Object-oriented programming language for application. smart contract development (0.8.11). ReactJS: JavaScript library for building user interfaces 3.2.4.1 Server configuration Amazon EC2 Mac (17.0.2). instances with x86-based architecture. Scikit-learn: Machine learning library for Python 8 vCPUs and 32 GB RAM allocated to handle the (0.24.1). computational demands of heart attack prediction PostgreSQL: Relational database management system models, including processing real-time data from (13.3). wearable devices and large-scale patient datasets. Python: Programming language used for backend 10 Gbps networking bandwidth for high-speed development (3.9.5). communication between server components, external services, and connected heart monitoring 3.3 Digital twin application devices. We developed a Digital Twin application model to moni- tor patients and predict brain strokes. This model utilizes 3.2.4.2 Storage configuration Amazon Elastic a machine learning algorithm to analyze historical data Block Store (EBS) with 100 GB capacity for storing for assessing the risk of brain strokes. While the focus of patient medical records, real-time heart monitoring this study is on brain stroke prediction, the prototype is data, and predictive model outputs. designed to be highly adaptable and can be extended to Total storage capacity of 30 GB allocated for data predict other conditions, such as heart attacks, cancers, management purposes. osteoporosis, and epilepsy. The architecture of the Digital Daily automated backups to store patient data Twin application is depicted in Fig. 1. and models securely, ensuring compliance with The process begins with data retrieval. According to the healthcare regulations and preventing data loss. guidelines from the Joint National Committee on Detec- tion, Evaluation, and Treatment of High Blood Pressure, 3.2.4.3 Software versions Ganache: Private hypertension levels should be validated approximately Ethereum blockchain environment (6.12.1). every month to reduce the risk of strokes. Following Fig. 1 Digital twin prototype (Architecture) Upadrista et al. Brain Informatics (2025) 12:1 Page 8 of 15 these guidelines, we have developed the application such data through a consensus mechanism, preventing tam- that patient data is retrieved from the Patient Data File pering before the data is finalized on the blockchain. every month, or whenever new readings are available (e.g., blood pressure or cholesterol levels). The Patient 3.3.2.2 Data storage The full patient data, along with its Data is stored in XML format to ensure easy integration corresponding hash, is stored in a PostgreSQL database. with future data sources, including wearable devices. Only the metadata, such as the hash keys and timestamps, Based on a predefined monthly schedule, the Digi- is stored on the blockchain. This ensures that while the talTwinUtility retrieves patient data from the Patient data integrity is maintained, the blockchain remains light- Data XML File. The data retrieved includes age, sex, past weight by not storing the entire dataset on-chain. The history of heart disease, brain stroke, work type, resi- hybrid approach allows for efficient data management dence type, smoking status, marital status, Body Mass while ensuring the security benefits of the blockchain. Index (BMI), hypertension, and average glucose levels. The data is then formatted and sent to the blockchain to 3.3.2.3 Tamper detection Each time the system retrieves ensure it is securely stored. The blockchain plays a crucial patient data for processing (e.g., predicting stroke risk), role in maintaining the security and accuracy of patient the hash stored in the PostgreSQL database is compared data. It generates a unique identifier for each data entry with the hash on the blockchain. Since the blockchain is and stores this along with the patient’s information in shared between the two nodes, any modification or tam- a secure database. Once the data is securely stored, the pering attempt will be immediately flagged by a mismatch machine learning model analyzes it to predict the risk in the hashes. The blockchain’s distributed nature ensures of a stroke. If the system predicts a high risk of stroke, that both nodes must agree on the data’s integrity, and any it automatically notifies healthcare providers via email. unauthorized changes will be rejected. Doctors can then use the system’s user-friendly interface to review the patient’s health data and prediction results. 3.3.2.4 Immutable ledger Due to the blockchain’s immu- The application’s flexible design makes it easy to adapt for table nature, once a transaction—comprising patient data predicting other illnesses and integrating new medical and its associated metadata—is recorded, it cannot be devices, making it a valuable tool in modern healthcare. altered or deleted. This ensures that a patient’s medical history is preserved in a secure and verifiable manner. 3.3.1 Logistic regression algorithm Any attempt to tamper with past data would break the The Logistic Regression algorithm has been used as part cryptographic chain of blocks, alerting both nodes and of this prototype because it is highly adaptable and can preventing fraudulent changes. be applied to other pathologies including heart attacks, By employing a consortium blockchain with two osteoporosis, epilepsy and similar conditions. Logistic nodes, we ensure decentralized control and data valida- Regression is a general-purpose classification algorithm tion, making it virtually impossible for a single entity to well-suited for binary outcomes, such as predicting tamper with patient records. The consensus mechanism whether a specific event (e.g., stroke, heart attack, or epi- between nodes enhances trust and ensures that patient leptic seizure) will occur. data is recorded accurately and securely. This approach, combined with our smart contract-based validation pro- 3.3.2 Consortium Blockchain for data integrity cess, ensures robust data integrity and security across the We employ a consortium blockchain model with two system. nodes, simulating a network of two hospitals, to secure patient data and ensure its integrity within our Digital 3.3.3 Application scalability Twin application. This blockchain model enhances secu- The architecture is adaptable to detecting various pathol- rity by preventing unauthorized access or tampering, ogies, such as heart attacks, cancers, osteoporosis or with each node sharing responsibility for data validation. epilepsy, by simply adjusting the features used for predic- The integration of the consortium blockchain within the tion. This adaptability enables the algorithm to be applied architecture is outlined below: across a broad range of conditions, with each requiring specific patient data to assess the likelihood of occur- 3.3.2.1 Smart contracts The application utilizes smart rence or disease progression. While Logistic regression is contracts, written in Solidity, to automate data integrity highly versatile for many binary classification problems, checks. Each time new patient data is added or updated, it has limitations when applied to certain medical pathol- the smart contract generates a unique hash (a crypto- ogies, particularly those with more complex relationships graphic fingerprint) for that data. This hash is then stored between variables or when the problem is not binary. For on the blockchain, ensuring data authenticity. Both nodes example, Parkinson’s Disease, may require a multiclass (representing hospitals) must agree on the validity of the classification model. In such cases, the application can Upadrista et al. Brain Informatics (2025) 12:1 Page 9 of 15 be reused with minimal modifications, such as updat- The SL (segment level) Slope is greater than 2 mm ing the machine learning model reference and adjust- (up), indicating significant ST-segment changes. ing the relevant patient data fields to fit the new model requirements. These parameters serve as major indicators for assessing The Digital Twin application also supports seamless the risk of a brain stroke. integration with wearable medical devices that provide We analyzed 62 patient records. Refer to Table 6 in the real-time patient data. Devices that monitor metrics like Appendix for a reference on the structure of individual blood pressure, glucose levels, and other vital signs can patient data. Two additional columns on the right side of automatically write this data to the Patient Data File. the table evaluate: the real risk of a patient having a brain Because the system architecture is loosely coupled, new stroke and the accuracy of the machine learning (ML) devices can be integrated without altering the core appli- model in predicting the risk of a brain stroke. cation code, ensuring adaptability and scalability across a To address potential biases in the dataset, we applied wide range of medical technologies. fairness evaluation techniques. Fairness evaluation tech- niques refer to methods used to assess the fairness of AI 3.3.4 Data visualization and user interface systems, particularly in terms of their impact on different Two user interfaces (Figs. 4) were developed to provide demographic groups, such as race, gender, age, or socio- healthcare providers with a clear view of patient data. economic status. These techniques aim to identify and These UIs allow doctors to: mitigate biases and discriminatory effects that AI systems may exhibit. Review patient health metrics and trends over time. We performed two additional experiments involv- Receive alerts about patients who are at high risk of ing different datasets categorized by patient age: one stroke. with patients aged over 70, another with patients aged Access detailed insights into the ML-predicted risk between 40 and 70, and a third with patients under 40. factors for each patient. The outcomes are presented in Table 5, revealing that the machine learning model effectively identified individuals 4 Results at risk of brain stroke with a prediction accuracy exceed- This section presents the results of our prototype, high- ing 98.28%. lighting its performance in terms of prediction accuracy, Based on the aggregation of all available results, the scalability, and security. Human Digital Twin (DT) model we developed has dem- onstrated a prediction accuracy of 98.28% in identifying 4.1 Digital twin application accuracy the risk of brain strokes. The likelihood of brain stroke in a patient is detected under the following conditions: 4.2 Digital twin application scalability The Digital Twin application uses a scalable Logistic Hypertension is identified (indicated by a value of 1). Regression model to predict the risk of brain stroke and The presence of heart disease is confirmed (indicated can be adapted to other critical illnesses, such as heart by a value of 1). attacks (by using features like cholesterol levels, blood Resting systolic blood pressure exceeds 180 mmHg. pressure, and family history) and epilepsy (using fea- Serum cholesterol level is greater than 240 mg/dL. tures such as BMI, blood sugar levels, and family history) Fasting blood glucose level falls within the range of allowing it to be effectively used for various conditions > 100 to 125 mg/dL, indicating prediabetes. with no code changes to the Digital twin application. The patient’s maximum heart rate exceeds the For pathologies that cannot use binary classification predicted maximum based on age. The predicted logistic regression model, there are just few steps, as maximum heart rate is determined by subtracting described below, to extending our digital twin applica- the patient’s age from 220. For example, a 35-year- tion to work for critical illness that requires complex old’s predicted maximum heart rate is 185 beats per computations. minute; exceeding this rate during physical activity is Add new fields (If required) –If a new illness requires considered risky. additional patient fields in the database, update the The OL (overshoot level) Peak exceeds 123. PatientData.xml file with the necessary fields. Table 5 Prediction results across multiple experiments Dataset Type Reference to the structure of individual patient data Prediction Accuracy % Masked Patient Data set aged > 70 Appendix A - Table 7 100% Masked Patient Data set aged < 70 and > 40 Appendix A - Table 8 98.28% Upadrista et al. Brain Informatics (2025) 12:1 Page 10 of 15 Modify the ML model reference – To configure the 4.3.1 Tampering detection via consortium blockchain application for predicting a different critical illness, In our consortium blockchain model, where two nodes update the model reference in the configuration.xml file (representing hospital organizations) share control of to point to the appropriate ML model for the new illness. the blockchain, such data tampering attempts are quickly Restart the application – After making the necessary identified. The blockchain operates through a consensus configuration changes, restart the application to activate mechanism, meaning both nodes must validate the integ- the prediction capabilities for the new illness. rity of the data before it is accepted into the ledger. In this This process highlights the Digital Twin application’s case, the tampered data caused a mismatch in the crypto- scalability in adopting to any other use cases. graphic hash, as the original values stored on the block- chain did not match the altered data in the PostgreSQL 4.2.1 Integration with other medical devices database. As depicted in Fig. 1, we have used CSV files to capture Upon detecting this discrepancy, the blockchain sys- patient real-time data such as resting average blood pres- tem immediately flagged the tampered transaction. The sure, average cholesterol level, average fasting blood smart contract automatically rejected the invalid data sugar levels, average resting electrocardiography, aver- and reverted the records back to their original state, as age maximum heart rate, exercise-induced angina, and validated by the hashes stored across both nodes of the hypertension levels. The application we have developed is consortium blockchain. loosely coupled from the patient data file (CSV files), as Additionally, as part of our alert system, an email noti- shown in Fig. 1 above. This setup enables seamless inte- fication (refer Fig. 2) was automatically triggered, inform- gration with other medical devices. The wearable devices ing the relevant administrators of the data tampering simply need to be configured to write the data directly attempt. The email highlighted the discrepancy and into the CSV files. No other changes are required to the reassured that the original values were retained across application, and the model we have developed will work the blockchain network, preventing any unauthorized seamlessly regardless of the diversity of medical devices changes from being committed. This real-time alert sys- that need to be integrated with the application. This fea- tem provides transparency and ensures that the stake- ture highlights the adaptability and scalability of our holders are promptly notified of any malicious activity application to integrate with medical devices. while maintaining data integrity. 4.3 Simulating and detecting data tampering in a 4.4 Data visualization consortium blockchain Within our Digital Twin application, doctors will pri- The security of the system and the health and safety of marily receive email notifications about a patient’s brain patients could be compromised in the event of a cyber- stroke risk. However, in some cases, they may want attack. To test the resilience of our consortium block- to review detailed real-time health data, including the chain setup, we simulated a data tampering attack by a patient’s medical history from electronic health records. malicious user targeting patient data stored in both the To address this, we have developed two alternative user Ganache consortium blockchain and the PostgreSQL interface flows (Figs. 3 and 4) that allow doctors to access database. In this simulation, we deliberately altered comprehensive medical reports for each patient. patient data to invalid values, mimicking a typical cyberattack. 4.4.1 Patient summary report The original data for Patient 6 included: The UI displaying all patient summary data is presented Body Mass Index (BMI) = 18. in Fig. 3, where the doctor can access records of all their Hypertension Level = 100. patients, including any associated brain stroke risks. Average Glucose Level = 100. These values indicated no immediate risk of brain 4.4.2 Detailed patient data stroke. In the tampering scenario, we changed the data The doctor can access the details of each patient data by from back end to: clicking on the individual patient’s name. Body Mass Index (BMI) = 146. Hypertension Level = 120. 4.5 Summary of results Average Glucose Level = 150. We have demonstrated the effectiveness of our prototype These altered values now implied that Patient 6 was at a in terms of prediction accuracy, scalability, and security. higher risk for brain stroke. Our application achieved a prediction accuracy of 98.28% in identifying the risk of brain strokes. This performance exceeds the accuracy levels reported in the literature, which typically range from 84 to 92% [9–15, 22–25]. Upadrista et al. Brain Informatics (2025) 12:1 Page 11 of 15 Fig. 2 Email Notification retaining the original values even though patient data was tampered from backend Fig. 3 UI 1 - Patients summary user interface Upadrista et al. Brain Informatics (2025) 12:1 Page 12 of 15 Fig. 4 UI 2 - Individual patient data Upadrista et al. Brain Informatics (2025) 12:1 Page 13 of 15 However, further rigorous validation is necessary to eval- the blockchain. Instead of storing all patient data on the uate our model’s performance against diverse, real-world blockchain, only patient metadata is stored, resulting in datasets, including those used in these literature studies. a lightweight and efficient blockchain implementation, Additionally, our model showcased superior scalabil- unlike other approaches in the literature where all data is ity, as it can be easily extended to predict other critical stored on the blockchain. illnesses like heart attacks, cancers, osteoporosis, and While the model has demonstrated an accuracy of epilepsy without requiring modifications to the core 98.28%, a limitation of this study is the use of a small and application code, which was not spoken about in the lit- curated dataset for brain stroke prediction and illustra- erature so far. tion of system’s scalability to other healthcare use cases. In terms of security, the integration of consortium Although the study highlights the accuracy, security, and blockchain technology ensures robust protection of scalability of the proposed system, additional experi- sensitive patient data. This approach goes beyond the ments and comparative analyses are necessary to sub- methods commonly employed in the literature, such as stantiate these advantages fully. This includes evaluating basic encryption techniques [10, 15, 23] access control the system against datasets referenced in existing litera- mechanisms [14, 22, 25] and data anonymization prac- ture to determine if it can consistently surpass the accu- tices [12, 27, 29] providing enhanced resilience against racy levels reported in prior studies. cyberattacks. In summary, the model presented in this paper can be used to address three key aspects: high predictive accu- 5 Discussion racy, robust security, and scalability for other pathologies Security and privacy concerns are significant challenges with minimal modifications to the application code. in healthcare [34–38], and many digital twin implemen- tations in the literature have not adequately addressed 6 Conclusion and future work these issues. Additionally, the models discussed in the This paper introduces a Digital Twin application inte- literature for detecting pathologies such as heart attacks grated with a machine learning model for predicting and brain strokes are often designed specifically for a pathologies such as heart attacks, cancers, osteoporosis single condition, limiting their ability to detect other and epilepsy. Unlike other models that focus on specific pathologies and restricting their scalability. Moreover, organs or single-use cases, our application replicates the the observed accuracy levels in the studies reviewed so entire human body, making it adaptable for predicting a far have yet to surpass the 92% accuracy threshold [9–11, wide range of medical conditions. Through minor con- 22–24]. figuration adjustments, the model can seamlessly predict In this paper, we present a Digital Twin application other diseases without the need for code modifications. designed to be easily extendable for various critical ill- Security and privacy are addressed comprehensively in nesses with minimal changes to the existing code. Ini- our model through the integration of consortium block- tially tested for brain stroke prediction using the logistic chain technology. This ensures robust protection of sen- regression algorithm, the application can be seamlessly sitive healthcare data, ensuring data. Additionally, our adapted for other conditions such as heart attacks, can- machine learning model achieved a prediction accuracy cers, osteoporosis, or epilepsy without further modifica- of 98.28%. Our application also features a loosely coupled tions. Additionally, it can handle more complex diseases architecture, allowing seamless integration with various where binary classification is insufficient, with detailed medical devices and ensuring compatibility with existing steps provided to ensure full scalability. The application healthcare infrastructures. can also be effortlessly expanded to integrate wearable Future work will validate the application performance devices, enabling the transmission of real-time patient using real-world datasets to ensure its robustness and data. The model has demonstrated an accuracy of 98.28% reliability in practical applications. We also plan to under controlled conditions. To address security and extend the Digital Twin model by integrating it with vari- privacy concerns, the model incorporates a Consortium ous wearable devices, such as blood pressure and heart blockchain-based solution, making it highly secure. We rate monitors. Additionally, we aim to enhance the appli- tested the application by deliberately tampering with cation’s security by implementing Advanced Encryp- backend data, and the system successfully detected the tion Standard (AES) across both the device layer and tampering and restored the original values. Smart con- the cloud platform. Given the blockchain’s decentralized tracts, written in Solidity, were employed to manage transactions higher latency than traditional databases, data validation and integrity checks across the nodes. we will consider employing off-chain solutions for faster These smart contracts automatically triggered correc- transactions, in conjunction with sharding techniques tive actions when discrepancies were found between to enable parallel processing across smaller blockchain the tampered data and the original metadata stored on segments. Upadrista et al. Brain Informatics (2025) 12:1 Page 14 of 15 Abbreviations References API Application Programming Interface 1. Vats T, Singh SK, Kumar S, Gupta BB, Gill SS, Arya V, Alhalabi W (2023) Explain- IoMT Internet of Medical Things able context-aware IoT framework using human digital twin for healthcare, RPM Remote Patient Monitoring Explainable Artificial Intelligence Solutions for In-the-wild Human Behavior XML Extensible Markup Language Analysis. EHR Electronic Health Record 2. Okegbile SD, Cai J, Yi C, Niyato D (2022) Human Digital Twin for Personalized ML Machine Learning Healthcare: Vision, Architecture and future directions. IEEE Network, pp. 1–7 ST Slope ST Segment Slope (used in cardiac monitoring) 3. Azzaoui AE, Kim TW, Loia V, Park J (2020) Blockchain-Based Secure Digital ECG Electrocardiography Twin Framework for Smart Healthy City. Adv Multimedia Ubiquitous Eng., Dec BMI Body Mass Index 4. Chakshu NK, Carson J, Sazonov I, Nithiarasu P (2019) A semi-active human GAN Generative Adversarial Network (not directly digital twin model for detecting severity of carotid stenoses from head vibra- mentioned but often abbreviated in related tion - a coupled computational mechanics and computer vision method: a contexts) semi-active human digital twin model for detecting carotid stenoses. Int J AES Advanced Encryption Standard Numer Methods Biomed Eng. OL Peak Overshoot Level Peak 5. He DD, Winokur ES, Sodini CG (2011) A continuous, wearable, and wireless SL Slope Segment Level Slope heart monitor using head ballistocardiogram (BCG) and head electrocardio- SPoF Single Point of Failure gram (ECG). In: 2011 Annual International Conference of the IEEE Engineering Ref. Reference number of the study in Medicine and Biology Society, pp. 4729–4732 Acc. Accuracy level of the model 6. Lombardo A, Ricci C (2022) Digital Twins Federation for Remote Medical Care Sec. Measures Security measures implemented of De-Hospitalized Patients. EEE International Conference on Pervasive Com- Encrypt. Encryption puting and Communications Workshops and other Affiliated Events (PerCom Access Ctrl. Access Control Workshops), pp. 718–723 Data Anonym. Data Anonymization 7. August-Wilhelm S (2024) The Composable Enterprise: agile, flexible, innova- Hom. Encryption Homomorphic Encryption tive. Springer Fachmedien Wiesbaden GmbH, p. 24 Proto. Sec. Prototype for Security/Privacy 8. Allen A, Siefkas A, Pellegrini E, Burdick H, Barnes G, Calvert J, Mao Q, Das Temp. Conv. Networks Temporal Convolutional Networks R (2021) A Digital Twins Machine Learning Model for forecasting Disease Geo., Temp. Geographical, Temporal Progression in Stroke patients. Appl Sci 11:5576. Feature Sel. Feature Selection 9. Smith J, Johnson A, Brown K (2022) Digital Twins for Brain Stroke Prediction. J Med Technol 45:123–136 Author contributions 10. Garcia M, Martinez L, Rodriguez P (2023) Enhanced Digital Twin models for All authors contributed to the study conception and design. Material Stroke Prediction. J Healthc Technol 12(3):45–57 preparation, data collection and analysis were performed by Venkatesh 11. Kim S, Lee H, Park J (2024) Deep learning-based Digital Twin Approach for Upadrista, Sajid Nazir and Huaglory Tianfield. The first draft of the manuscript Stroke Prediction. J Med Inf 36:78–92 was written by Venkatesh Upadrista and all authors commented on 12. Gonzalez R, Lopez M, Hernandez S (2023) Fuzzy logic-based Digital Twin previous versions of the manuscript. All authors read and approved the final Approach for Stroke Risk Prediction. J Med Inf 36(2):78–91 manuscript. 13. Martinez J, Lopez A, Perez G (2024) Temporal Convolutional Networks for Stroke Prediction in Digital Twins. J Biomed Inform 28(4):56–69 Funding 14. Wang H, Zhang L, Liu Y (2023) Dynamic Bayesian Network Approach for No funding was received for conducting this study. Digital Twin-based stroke prediction. J Healthc Inf Res 10(2):45–57 15. Rodriguez M, Perez C, Sanchez J (2023) Genetic algorithm-based Digital Twin Data availability Approach for Stroke Risk Prediction. J Med Inform 35(4):112-25. The datasets generated during and/or analyzed during the current study are 16. Mamta BB, Gupta K-C, Li VCM, Leung, Psannis KE (2021) Blockchain-assisted retrieved from Kaggle https://www.kaggle.com/datasets/jillanisofttech/brai secure fine-grained searchable encryption for a cloud-based Healthcare n-Stroke-dataset. Kaggle is one of the most reliable and authentic platforms Cyber-physical System. IEEE/CAA J Automatica Sinica, pp. 1877–1890 for data scientists and researchers, recognized by reputable sources such as 17. Dai H-N, Imran M, Haider N (2020) Blockchain-Enabled Internet of Medical Forbes. We have created real time patient data which constitute of patient Things to Combat COVID-19IEEE Internet Things Mag 3:52-7 individual readings (e.g., blood pressure, Heart rate, brain related readings) 18. Wu S, Du J (2019) Electronic medical record security sharing model based on which can be downloaded from https://github.com/vupadrista/Blockchai blockchain, Proceedings of the 3rd International Conference on Cryptogra- n-Enabled-Digital-Twin-System-for-Brain-Stroke-Prediction. All appendices phy, Security and Privacy, pp. 13–17 referenced in this paper are available at https://figshare.com/articles/journal_c 19. Raj A, Prakash S (2022) A Privacy-Preserving Authentic Healthcare Monitoring ontribution/Blockchain-Enabled_Digital_Twin_System_for_Brain_Stroke_Pred System Using Blockchain. International Journal of Software Science and iction/27980363?file=51027824 Computational Intelligence, pp. 1–23, International 20. Wang R, He J, Liu C, Li Q, Tsai W-T, Deng E (2018) A Privacy-Aware PKI System Code availability Based on Permissioned Blockchains, IEEE 9th International Conference on The code used during and/or analysed during the current study is available at Software Engineering and Service Science (ICSESS), pp. 928–931 https://github.com/vupadrista/Blockchain-Enabled-Digital-Twin-System-for-B 21. Zheng P, Xu Q, Zheng Z, Zhou Z, Yan Y, Zhang H (2022) Meepo: multiple rain-Stroke-Prediction. execution environments per Organization in Sharded Consortium Block- chain. IEEE J Sel Areas Commun 40:12 22. Chen Y, Wang Q, Liu X (2023) Personalized Digital Twin for Stroke Risk Assess- Declarations ment. J Healthc Eng Vols 9(4):112–125 23. Patel R, Sharma A, Gupta S (2024) Hybrid Digital Twin Framework for early Ethical approval stroke detection. J Biomed Eng 16(2):34–47 This work does not involve the use of human subjects. 24. Nguyen T, Tran L, Pham T (2023) Real-time Digital Twin Simulation for Stroke Prediction. J Med Simul Vols 8(1):56–67 Competing interests 25. Brown D, Miller K, Wilson E (2022) Predictive modeling of stroke events using The authors declare no competing interests. Digital twins. J Healthc Inf Vols 14(3):112–125 26. Liu W, Wang X, Zhang M (2024) Hybrid machine learning and physiological Received: 27 June 2024 / Accepted: 27 December 2024 model-based Digital Twin Framework for Stroke Prediction. J Healthc Eng Vols 11(1):78–91 27. Garcia J, Martinez A, Lopez M (2023) Spatiotemporal Digital Twin Approach for Stroke Prediction. J Biomed Inform 29(3):56–69 Upadrista et al. Brain Informatics (2025) 12:1 Page 15 of 15 28. Chen W, Wang T, Zhang L (2024) Multimodal Data Fusion Approach for Stroke 37. Garg N, Wazid M, Das AK, Singh DP, Rodrigues JJPC, Park Y (2020) BAKMP- Prediction using Digital twins. J Healthc Inf Res Vols 11(4):45–57 IoMT: design of Blockchain enabled Authenticated Key Management Proto- 29. Kim M, Lee J, Park E (2023) Feature selection-based Digital Twin Approach for col for Internet of Medical things Deployment. IEEE Access 8:95956–95977 Stroke Risk Prediction. J Med Simul Vols 9(2):45–57 38. Zaman S, Khandaker MRA, Khan RT, Tariq F, Wong K-K (2021) Thinking out of 30. Martinez D, Lopez C, Garcia A (2024) Dynamic updating mechanism for the blocks: Holochain for distributed security in IoT Healthcare. IEEE Internt Digital Twins in Stroke Prediction. J Med Technol 47(1):78–91 Things J 31. Wang X, Zhang W, Liu P (2023) Causality-based Digital Twin Approach for 39. Forebs KA To Connect Companies With Successful Data Scientists, Forebs, understanding stroke occurrence. J Healthc Technol 14(2):34–47 [Online]. Available: https://www.forbes.com/sites/alexknapp/2013/03/06/kag 32. TECH JS (2022) Brain Stroke Dataset, Kaggle, [Online]. Available: https://www. gle-aims-to-connect-companies-with-successful-data-scientists/ kaggle.com/datasets/jillanisofttech/brain-stroke-dataset 33. Pedelty L, Gorelick P (2004) Chronic Management of Blood Pressure After Stroke. Hypertension. Publisher’s note 34. Alrawais A, Alhothaily A, Hu C, Cheng X (2017) Fog computing for the inter- Springer Nature remains neutral with regard to jurisdictional claims in net of things: security and privacy issues. IEEE Internet Comput 21:2 published maps and institutional affiliations. 35. Gai K, Qiu M, Xiong Z, Liu M (2018) Privacy-preserving multichannel com- munication in edge-of-things. Future Generation Comput Syst 85:190–200 36. Ray PP, Dash D, Salah K, Kumar N (2021) Blockchain for IoT-Based Healthcare: background, Consensus, platforms, and use cases. IEEE Syst J (Early Access), pp. 85–94

Blockchain-Enabled Digital Twin System for Brain Stroke Prediction PDF

Document Details

Tags

Related

Summary

Full Transcript