CoChat

Blockchain-Enabled Trust and Compliance for Clinical AI: Decentralized Governance Without Decentralized Data Storage

Open access

The DOI matches its Crossref record; 76 of the 76 references it cites were found in a registry or an index; none of the 26 sources its claims cite is checked for support yet.

Abstract

Clinical AI systems rely on evolving machine learning pipelines and large-scale medical imaging data, creating persistent challenges in trust, auditability, consent governance, and reproducibility. This paper proposes a decentralized governance framework for clinical AI that uses blockchain as a verification and policy-enforcement overlay without decentralizing sensitive medical data storage or clinical inference. Raw images and clinical artifacts remain in secure repositories, while cryptographic commitments, consent states, access events, and reproducibility manifests are anchored to a tamper-evident ledger. The framework enables verifiable provenance, programmable consent enforcement, auditable execution, and deterministic reconstruction of AI-assisted decisions while preserving regulatory alignment and clinical usability. In a medical imaging proof-of-concept spanning nine simulated scenarios and approximately 43,500 inference attempts across cohorts of 50 to 1000 subjects, the framework achieved a mean Governance Quality Index of 0.93, governance overhead below 11 ms per operation under routine settings, and throughput above 220 requests per second on average. Complementary validation over a real-world paired imaging-and-clinical dataset structure further showed that the same governance abstractions can be instantiated for 625 subjects and 6349 MR images. Overall, the framework separates governance from data and computation, providing verifiable auditability and reproducibility without disrupting existing clinical infrastructures.

Overview

CoChat Summary

TL;DR

The paper proposes a blockchain-based governance overlay for clinical AI that anchors consent, audit events, and reproducibility manifests on a tamper-evident ledger while keeping medical data and inference off-ledger. In a simulated dermoscopic imaging proof-of-concept across nine scenarios and approximately 43,500 inference attempts, it reports a mean Governance Quality Index of 0.927, mean governance latency of 10.54 ms, and average throughput of 224.27 requests/s.

Key findings

  • The composite Governance Quality Index (GQI) achieved a mean of 0.927 (95% CI: [0.880, 0.975]); security and compliance components each had a mean of 1.000, performance 0.850, and auditability 0.667 (SD 0.500). Six of nine scenarios (67%) met the production-readiness threshold (GQI ≥ 0.90), and the remaining three were classified as acceptable.
  • Across all scenarios, mean governance latency was 10.54 ms per operation, median latency was 4.39 ms, and the 95th percentile was 30.29 ms; average throughput exceeded 224 requests/s, with peak throughput near 798 requests/s in small-scale settings. In ablation, as patient population increased from 50 to 1000, mean per-request governance latency rose from 1.25 ms to 42.9 ms and throughput fell from 798.2 RPS to 23.3 RPS.
  • Hash-chain verification detected all injected ledger tampering attempts, and off-ledger manifest integrity checks achieved 100% sensitivity, 100% specificity, and zero false alarms, correctly detecting all 50 tampered manifests while verifying all 150 legitimate manifests. Mean audit completeness was 1.169, exceeding 100% of anticipated events.
  • Complementary validation over a real-world paired imaging-and-clinical dataset structure showed the same governance abstractions could be instantiated for 625 subjects and 6349 MR images.

Methods

  • Design: A layered decentralized governance overlay for clinical AI, where the ledger records cryptographic commitments, consent states, access events, and reproducibility manifests; raw images and clinical artifacts remain off-ledger, and no AI inference is performed on-chain. The GQI is a weighted linear combination of security (w=0.35), compliance (w=0.35), performance (w=0.15), and auditability (w=0.15).
  • Data or sample: A medical imaging proof-of-concept using simulated dermoscopic image analysis across nine scenarios varying patient scale (50–1000), inference runs (500–20,000), consent revocation probability (0.00–0.30), and tampering (yes/no), totaling approximately 43,500 inference attempts. A complementary real-world paired imaging-and-clinical dataset structure included 625 subjects and 6349 MR images.
  • Main analysis: The prototype ledger, LocalLedger, was implemented in Python 3.11+ as a single-node, file-based append-only ledger using JSON Lines records, SHA-256 hash chaining, and deterministic serialization. Evaluation combined scenario-based simulation, descriptive statistics and confidence intervals, security testing for authorization and integrity, and ablation studies; algorithms formalize request-time governance enforcement, ledger integrity verification, and off-ledger manifest verification.

Limitations

  • The paper states that the experimental evaluation is based on a limited set of simulated scenarios that primarily explore monotonic scaling of patient population, workload intensity, and consent revocation probability, and does not capture bursty workloads, multi-institutional access patterns, or heterogeneous mixes of authorized and unauthorized requests.
  • The paper states that adversarial fault-injection scenarios are evaluated alongside normal-operation scenarios, and future work should separate normal and adversarial evaluation tracks; it also states that the current evaluation does not fully model high-concurrency conditions, long-term longitudinal operation, or geographically distributed deployments.
  • The paper states that the prototype does not include distributed consensus, Byzantine fault tolerance, or formal protocol verification, and that deployment would require integration with PACS/VNA, EHR, DICOM/FHIR workflows, identity federation, key management, policy administration, audit export, operational monitoring, and institutional approval. The text also makes evident that the single-node LocalLedger measurements are an empirical lower-bound baseline rather than a distributed-ledger deployment benchmark.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/electronics15143046

Cites

76 of 76 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Overview of the proposed blockchain-enabled governance framework for clinical AI. Conventional clinical workflows-including image acquisition, AI analysis, and clinical reporting-remain unchanged, while a blockchain-backed governance overlay enforces consent verification, records reproducibility manifests, and anchors immutable audit events without storing medical data or performing AI inference on-chain. A complementary decentralized computation workflow supports privacy-preserving training and experimentation on de-identified or synthetic datasets. Together, these mechanisms enable explainable AI, reproducible and auditable clinical decisions, consent-aware execution, and quantitative governance evaluation through the Governance Quality Index (GQI), demonstrating production-ready governance (GQI ≥ 0.90).
Figure 1 Overview of the proposed blockchain-enabled governance framework for clinical AI. Conventional clinical workflows-including image acquisition, AI analysis, and clinical reporting-remain unchanged, while a blockchain-backed governance overlay enforces consent verification, records reproducibility manifests, and anchors immutable audit events without storing medical data or performing AI inference on-chain. A complementary decentralized computation workflow supports privacy-preserving training and experimentation on de-identified or synthetic datasets. Together, these mechanisms enable explainable AI, reproducible and auditable clinical decisions, consent-aware execution, and quantitative governance evaluation through the Governance Quality Index (GQI), demonstrating production-ready governance (GQI ≥ 0.90).CC-BY · From the PDF
Figure 2 Governance performance and scalability characteristics across evaluated scenarios. ( Top ) Distribution of per-request governance latency, showing that median overhead remains below the real-time target of 10 ms and all scenarios remain within the acceptable clinical threshold of 50 ms. ( Bottom left ) Governance overhead as a function of workload size, demonstrating near-linear scaling with increasing request volume and patient population. ( Bottom right ) Distribution of system throughput across scenarios, with a mean throughput of 224.3 requests/s exceeding the nominal clinical target of 100 requests/s, indicating that governance enforcement does not compromise operational capacity.
Figure 2 Governance performance and scalability characteristics across evaluated scenarios. ( Top ) Distribution of per-request governance latency, showing that median overhead remains below the real-time target of 10 ms and all scenarios remain within the acceptable clinical threshold of 50 ms. ( Bottom left ) Governance overhead as a function of workload size, demonstrating near-linear scaling with increasing request volume and patient population. ( Bottom right ) Distribution of system throughput across scenarios, with a mean throughput of 224.3 requests/s exceeding the nominal clinical target of 100 requests/s, indicating that governance enforcement does not compromise operational capacity.CC-BY · From the PDF
Figure 3 Auditability and integrity verification of the governance ledger across evaluated scenarios. ( Left ) Audit trail completeness, measured as the ratio of actual ledger entries to expected governance events, showing that all scenarios exceed the 100% completeness baseline, with a mean completeness of 1.169, reflecting additional governance and control events beyond core inference execution. ( Right ) Ledger integrity verification using cryptographic hash-chain validation, where all scenarios pass pre-test verification, while post-test verification correctly detects integrity violations in intentionally tampered scenarios (3/9), demonstrating the tamper-evident properties of the governance ledger rather than indicating system failure
Figure 3 Auditability and integrity verification of the governance ledger across evaluated scenarios. ( Left ) Audit trail completeness, measured as the ratio of actual ledger entries to expected governance events, showing that all scenarios exceed the 100% completeness baseline, with a mean completeness of 1.169, reflecting additional governance and control events beyond core inference execution. ( Right ) Ledger integrity verification using cryptographic hash-chain validation, where all scenarios pass pre-test verification, while post-test verification correctly detects integrity violations in intentionally tampered scenarios (3/9), demonstrating the tamper-evident properties of the governance ledger rather than indicating system failureCC-BY · From the PDF
Figure 4 Off-ledger manifest tampering detection performance. ( Top left ) Confusion matrix summarizing tampering detection outcomes across 200 tested manifests, showing perfect separation between tampered and legitimate artifacts. ( Top right ) Detection performance metrics, where sensitivity, specificity,
Figure 4 Off-ledger manifest tampering detection performance. ( Top left ) Confusion matrix summarizing tampering detection outcomes across 200 tested manifests, showing perfect separation between tampered and legitimate artifacts. ( Top right ) Detection performance metrics, where sensitivity, specificity,CC-BY · From the PDF
Image 5
CC-BY · From the PDF
Figure 6 Ablation study of governance overhead and deployment readiness under varying operational parameters. ( Top left ) Impact of patient population size on average governance overhead, showing near-linear scaling with increasing patient counts. ( Top right ) Impact of workload intensity (number of inference runs), demonstrating stable and predictable overhead growth even under high-throughput conditions. ( Bottom left ) Impact of consent revocation probability, illustrating increased overhead under high revocation churn while maintaining perfect consent compliance. ( Bottom right ) Mean Governance Quality Index (GQI) across all ablation parameters, indicating that most configurations meet or exceed the production-readiness threshold, with lower scores driven primarily by auditability effects in high-stress or adversarial conditions.
Figure 6 Ablation study of governance overhead and deployment readiness under varying operational parameters. ( Top left ) Impact of patient population size on average governance overhead, showing near-linear scaling with increasing patient counts. ( Top right ) Impact of workload intensity (number of inference runs), demonstrating stable and predictable overhead growth even under high-throughput conditions. ( Bottom left ) Impact of consent revocation probability, illustrating increased overhead under high revocation churn while maintaining perfect consent compliance. ( Bottom right ) Mean Governance Quality Index (GQI) across all ablation parameters, indicating that most configurations meet or exceed the production-readiness threshold, with lower scores driven primarily by auditability effects in high-stress or adversarial conditions.CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 2 Not valid · 17 Not checked · 197 Own results and interpretations

48 more claims

Create a free account to see all 216 claims and to verify the 25 references still open.

See all claims — free account
  • The quote only partly states this claim. 1
  • It may state two separate claims as one. 94
  • Probably not a substantial claim. 113

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 76 references

Research works printed but not linked: 3

  1. Panagoulias, D.P.; Tsihrintzis, G.A.; Virvou, M. Challenges in Regulating and Validating AI-Driven Healthcare. In Artificial Intelligence-Empowered Bio-medical Applications; Springer: Berlin/Heidelberg, Germany, 2025; pp. 135-152. · DOI
  2. McMahan, B.; Moore, E.; Ramage, D.; Hampson, S.; y Arcas, B.A. Communication-efficient learning of deep networks from decentralized data. In Proceedings of the Artificial Intelligence and Statistics; PMLR: New York, NY, USA, 2017; pp. 1273-1282.
  3. Zaharia, M.; Chen, A.; Davidson, A.; Ghodsi, A.; Hong, S.A.; Konwinski, A.; Murching, S.; Nykodym, T.; Ogilvie, P.; Parkhe, M.; et al. Accelerating the machine learning lifecycle with MLflow. IEEE Data Eng. Bull. 2018, 41, 39-45.

Metadata

Authors
Panagoulias, D. P., Damian, A. I., Stamate, C., Toderian, V., Butusina, P., De Franceschi, A., Bleotiu, C., Sakkopoulos, E., & Tsichrintzi, E.-A.
Affiliations
  • University of Piraeus (GR)
  • Universitatea Națională de Știință și Tehnologie Politehnica București (RO)
  • Birkbeck, University of London (GB)
  • Ratio1.ai, 031681 Bucharest, Romania
  • Noetiv PC, Valaoritou 18, 106 71 Athens, Greece
Venue
Electronics
Publisher
MDPI AG
Published
2026-07-10
Volume
15
Issue
14
Pages
3046
ISSN
2079-9292 (ISSN-L)
OpenAlex
W7167886567
Cited by
1
References
75
Citations by year
2026: 1
FWCI
2.31
Type
journal article
Language
en
Open access
Locations
  • Electronics (publishedVersion, cc-by)
  • Preprints.org (submittedVersion, cc-by)
Funders
University of Piraeus Research Centre
Topics
  • Physical Sciences › Computer Science › Information Systems › Blockchain Technology Applications and Security
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Information Systems › Blockchain Technology in Education and Learning
Keywords
blockchain, decentralized governance, diagnostic imaging, data provenance, consent management, auditability, reproducibility, cryptographic commitments, regulatory compliance, MRI images

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-22, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-22 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.