Metraipolis › archive › dc_53152ca8189f
MCP-MedImg: Enabling Standardized PACS-AI Integration through the Model Context Protocol (2026) · DOI 10.21203/rs.3.rs-11357800/v1
- Source
- Metraipolis archive
- Original
- https://doi.org/10.21203/rs.3.rs-11357800/v1
- License (as archived)
- CC-BY-4.0
- Archived
- 2026-10-18T13:44:38.013Z
- SHA-256 of the archived text
6f249573659862e8d4aef9e4ec70c429ca6822344d52074aa30d37a3faabe9bf- Stable URL of this page
https://metraipolis.com/documento/dc_53152ca8189f
How to cite
MCP-MedImg: Enabling Standardized PACS-AI Integration through the Model Context Protocol (2026) · DOI 10.21203/rs.3.rs-11357800/v1 — Metraipolis archive, https://doi.org/10.21203/rs.3.rs-11357800/v1 (CC-BY-4.0); archived copy: https://metraipolis.com/documento/dc_53152ca8189f (Metraipolis archive dc_53152ca8189f, sha256 6f249573659862e8d4aef9e4ec70c429ca6822344d52074aa30d37a3faabe9bf)
BibTeX
@misc{dc53152ca8189f,
title = {MCP-MedImg: Enabling Standardized PACS-AI Integration through the Model Context Protocol (2026) · DOI 10.21203/rs.3.rs-11357800/v1},
howpublished = {Metraipolis archive},
url = {https://doi.org/10.21203/rs.3.rs-11357800/v1},
note = {License CC-BY-4.0; archived copy https://metraipolis.com/documento/dc_53152ca8189f; sha256 6f249573659862e8d4aef9e4ec70c429ca6822344d52074aa30d37a3faabe9bf},
year = {2026}
}Archived text
Reproduced under CC-BY-4.0. Attribution: Metraipolis archive, https://doi.org/10.21203/rs.3.rs-11357800/v1. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Articolo recente con licenza aperta: scheda dai metadati Crossref (posted-content). Titolo: MCP-MedImg: Enabling Standardized PACS-AI Integration through the Model Context Protocol Autori: xiong shi Pubblicato: 2026-10-14 · editore: Springer Science and Business Media LLC DOI: https://doi.org/10.21203/rs.3.rs-11357800/v1 Licenza dell'articolo: CC-BY-4.0 Abstract (testo degli autori, CC-BY-4.0): Abstract Background: The proliferation of AI-assisted diagnostic models in medical imaging has created a fragmented integration landscape with PACS systems, leading to redundant development, inconsistent security compliance, and disconnected clinical workflows. Most AI diagnostic products exist as standalone Software as a Medical Device (SaMD), limiting scalability, auditability, and clinical adoption. Existing integration solutions (Direct Plugin, DICOM Router, VNA) are point-to-point adaptations without unified semantic layers. While the Model Context Protocol (MCP) provides tool invocation frameworks for AI agents, it lacks dedicated extensions for the medical imaging domain. Critically, global regulators (FDA, NMPA, EU) are converging on total product life cycle (TPLC) approaches demanding continuous post-market surveillance, yet no standardized protocol supports such surveillance at scale. Objective: We propose MCP-MedImg, an open reference architecture and protocol specification for interoperable integration and governance of AI services in medical imaging ecosystems, allowing AI capabilities to be discovered, invoked, monitored, and governed across heterogeneous healthcare environments without replacing existing PACS infrastructures. Methods: MCP-MedImg extends MCP with a four-layer architecture: Transport (DICOMweb-compatible, with dicom-mcp upward compatibility), Security and Compliance (PHI de-identification plus 14-column audit trail), AI Orchestration (model routing and version management), and Application Semantics (inspection-driven routing, lesion encoding, conditional human-in-the-loop verification). The specification defines six core information model entities and a tiered conformance model (five mandatory requirements, seven optional extensions). A reference implementation based on Orthanc PACS and nnU-Net validates protocol consistency, workflow, multi-model routing, DICOM SR/SEG generation, and audit trails. Results: The reference implementation demonstrated end-to-end interoperability, standardized service registration, structured result integration (DICOM SR/SEG), and auditable tracking of AI lifecycle events. On an NVIDIA RTX 4090 D, end-to-end latency under realistic clinical conditions was: LIDC-IDRI lung nodule CT (200 volumes, mean 5.4 s, P99 22.6 s) and MSD Heart cardiac MRI (10 volumes, mean 18.2 s, P99 27.6 s, 5-fold ensemble Dice 0.9324). Validation comprised 132 passing tests across 10 verification scenarios. Conclusions: MCP-MedImg aims to transform AI-assisted diagnosis from standalone software into auditable, service-oriented capabilities with conditional human-in-the-loop verification—directly addressing the regulatory surveillance gap and establishing foundations for governable public infrastructure in healthcare AI. By defining a tiered conformance model, the protocol enables incremental adoption: from minimal compliance (Level 1) through full production deployment (Level 3). The service-oriented delivery mo... Fonte: https://api.crossref.org/works/10.21203%2Frs.3.rs-11357800%2Fv1 — metadati Crossref, consultati il 2026-10-18. Interrogazione: «model context protocol».