Edge computing and blockchain for privacy-critical and data-sensitive health applications
Nawaz, Anum (2025-02-23)
Edge computing and blockchain for privacy-critical and data-sensitive health applications
(23.02.2025)
Turun yliopisto
Julkaisun pysyvä osoite on:
https://urn.fi/URN:ISBN:978-952-02-0082-4
https://urn.fi/URN:ISBN:978-952-02-0082-4
Tiivistelmä
The widespread adoption of ubiquitous healthcare enhances the accessibility, quality, and efficiency of healthcare; however, the risk of data privacy breaches is also increasing due to third party service providers. According to Cisco’s annual report (2018–2023), 94% data processing on cloud servers and continuous tracing of personal data create privacy vulnerabilities. To deal data privacy challenges, recent studies integrate edge computing with distributed ledger-based technologies (DLTs) such as blockchain, but optimization of privacy requirements is still needed. Therefore, this study is conducted to minimize the existing gaps related to data privacy by proposing edge-intelligence (lightweight machine learning/deep learning) based distributed system EdgeBot. Specifically, our framework aims to optimize data privacy, fine-grained access control, and data ownership rights of real-time healthcare systems in time sensitive scenarios.
Initially, we proposed computing model of EdgeBot, formulate on/off chain secure communication and fine-grained data access scheme for processed health monitoring data. Thereafter, we constructed resource-efficient one directional convolutional neural network (1D-CNN) for multiclass arrhythmia detection in real-time health monitoring system. Bayesian optimization algorithm is embedded within the network to adaptively select the optimal combination of hyperparameters. We employed 2-channel ECG system using AD8232 along with STM32F427 and raspberry pi boards as edge gateways. To ascertain the precision and credibility of AD8232 based ECG system, a comparison of HR and RR interval measurement was obtained from Polysomnography (PSG) device. Comparative analysis of our proposed 1DCNN shows average of 97.4% accuracy while utilizing significantly fewer resources. Average ECG processing time along with data sharing on private ethereum requires only 150ms and 143ms respectively.
Secondly, P2P trustless data trade system is formulated utilizing edge gateways (pi 3, model B+), and STM32F427 (M3,M4 and M7) as lightweight nodes, leveraging ethereum. Results of employing ECDSA and ECIES on M3,M4 and M7 shows an average of 17.253s, 1.462s and 1.156s execution time respectively, while average power consumption by M3 and M4 was 200mW, whereas M7 uses an average of 290mW. Results indicate the superior performance of M4 cortex microcontrollers while consuming less resources. Resource and performance analysis shows less than 40% of average computing resources were utilized during transaction handling; it is remarkable that TRD requires only 34.6ms, VTR 36ms, and TCT 73.6ms on averiv age. FobSim simulator was utilized to check the scalability of the proposed scheme and results shows that gossip protocol decreases 35% latency during parallel transactions ranging from 5 to 500. EdgeBot post-quantum resistant extension was implemented, and results were compared with other lightweight KEMs. Latency and memory efficiency of Kyber512 KEM with other lightweight KEMs was compared on STM32F427 and it required 22-25 bytes of memory. Moreover, Kyber512 was found to be the best performer, balancing energy consumption and memory usage. Results of performance analysis shows EdgeBot, a viable option for fortifying data trade, data ownership, and exchange through edge gateways.
Thirdly, we presented the implementation and evaluation of real-time data sharing and tracing among trusted stakeholders while preserving transparency using sawtooth on a linux platform, leveraging AWS EC2 instances for server communication. We propose service optimization in sawtooth, a mathematical foundation for determining the most efficient combination of databatch size, transactions per second (tps), resource utilization and network resources while ensuring the reliability of transaction commitment. Furthermore, performance evaluations were conducted on both AWS and local PCs, utilized cAdvisor for docker containers and cloudWatch for AWS metrics. Results indicated significant spikes in CPU and network usage during transaction processing, with the system successfully managing 82% of 1,000 parallel statements. Overall, results analysis shows high scalability and reliability of EdgeBot while utilizing less resources as compared to recent studies, particularly for large-scale operations and seamless integration with diverse underlying DLTs.
Initially, we proposed computing model of EdgeBot, formulate on/off chain secure communication and fine-grained data access scheme for processed health monitoring data. Thereafter, we constructed resource-efficient one directional convolutional neural network (1D-CNN) for multiclass arrhythmia detection in real-time health monitoring system. Bayesian optimization algorithm is embedded within the network to adaptively select the optimal combination of hyperparameters. We employed 2-channel ECG system using AD8232 along with STM32F427 and raspberry pi boards as edge gateways. To ascertain the precision and credibility of AD8232 based ECG system, a comparison of HR and RR interval measurement was obtained from Polysomnography (PSG) device. Comparative analysis of our proposed 1DCNN shows average of 97.4% accuracy while utilizing significantly fewer resources. Average ECG processing time along with data sharing on private ethereum requires only 150ms and 143ms respectively.
Secondly, P2P trustless data trade system is formulated utilizing edge gateways (pi 3, model B+), and STM32F427 (M3,M4 and M7) as lightweight nodes, leveraging ethereum. Results of employing ECDSA and ECIES on M3,M4 and M7 shows an average of 17.253s, 1.462s and 1.156s execution time respectively, while average power consumption by M3 and M4 was 200mW, whereas M7 uses an average of 290mW. Results indicate the superior performance of M4 cortex microcontrollers while consuming less resources. Resource and performance analysis shows less than 40% of average computing resources were utilized during transaction handling; it is remarkable that TRD requires only 34.6ms, VTR 36ms, and TCT 73.6ms on averiv age. FobSim simulator was utilized to check the scalability of the proposed scheme and results shows that gossip protocol decreases 35% latency during parallel transactions ranging from 5 to 500. EdgeBot post-quantum resistant extension was implemented, and results were compared with other lightweight KEMs. Latency and memory efficiency of Kyber512 KEM with other lightweight KEMs was compared on STM32F427 and it required 22-25 bytes of memory. Moreover, Kyber512 was found to be the best performer, balancing energy consumption and memory usage. Results of performance analysis shows EdgeBot, a viable option for fortifying data trade, data ownership, and exchange through edge gateways.
Thirdly, we presented the implementation and evaluation of real-time data sharing and tracing among trusted stakeholders while preserving transparency using sawtooth on a linux platform, leveraging AWS EC2 instances for server communication. We propose service optimization in sawtooth, a mathematical foundation for determining the most efficient combination of databatch size, transactions per second (tps), resource utilization and network resources while ensuring the reliability of transaction commitment. Furthermore, performance evaluations were conducted on both AWS and local PCs, utilized cAdvisor for docker containers and cloudWatch for AWS metrics. Results indicated significant spikes in CPU and network usage during transaction processing, with the system successfully managing 82% of 1,000 parallel statements. Overall, results analysis shows high scalability and reliability of EdgeBot while utilizing less resources as compared to recent studies, particularly for large-scale operations and seamless integration with diverse underlying DLTs.
Kokoelmat
- Väitöskirjat [2871]