Tue, Oct 27, 2026
[Archive]
Volume 18, Issue 2 (June-2026 2026)
Iranian Journal of Blood and Cancer 2026, 18(2): 107-120 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Babaei S, Ghaderzadeh M, Salehnasab C. From Data to Decision: A Conceptual Architecture of Agentic AI for Hemato-Oncology Tumor Boards. , Iranian Journal of Blood and Cancer. 2026; 18 (2) :107-120
URL: http://ijbc.ir/article-1-1925-en.html
URL: http://ijbc.ir/article-1-1925-en.html
1- Department of Nursing, Naghadeh School of Nursing, Urmia University of Medical Sciences, Urmia, Iran.
2- Boukan Faculty of Medical Sciences, Urmia University of Medical Sciences, Urmia, Iran. ,mustafa.ghaderzadeh@gmail.com
3- Social Determinants of Health Research Center, Yasuj University of Medical Sciences, Yasuj, Iran.
2- Boukan Faculty of Medical Sciences, Urmia University of Medical Sciences, Urmia, Iran. ,
3- Social Determinants of Health Research Center, Yasuj University of Medical Sciences, Yasuj, Iran.
Abstract: (20 Views)
Background and Objective: Hematological malignancies are facing increasing complexity in diagnosis and treatment, which requires multidisciplinary decision-making in tumor boards. Traditional predictive AI models have limited clinical application due to limitations such as black-box nature and inability to reason in multiple steps. This study aimed to design a comprehensive conceptual framework for a multi-agent Agentic AI system architecture to support decision-making in hemato-oncology tumor boards.
Methods: This conceptual study was conducted using the Design Science Research methodological framework. The proposed architecture was developed based on the convergence of three reference frameworks, including the ReAct reasoning paradigm, the HL7 FHIR R4 standard, and the NIST AI RMF 1.1 framework. The system architecture was described using Unified Modeling Language (UML) 2.5, and four specialized agents (Data Parser, Genomics, Guideline RAG, and Orchestrator) were specified based on a standard seven-component design pattern. The architecture was evaluated using the multidimensional RAMSES framework by an eight-member multidisciplinary expert panel.
Results: The proposed architecture was designed with five distinct layers (data acquisition, agent coordination, knowledge and tools, safety and governance, and presentation). The system employed a multilayered Defense-in-Depth security approach with four independent protective layers, which reduced the hallucination rate from 18.4% to 3.2%. An application scenario based on an AML case showed that the system was able to process a complex clinical file in 3 minutes and 21 seconds and produce a 2,847-word structured report with a confidence score of 0.82. The RAMSES evaluation yielded an overall average score of 4.4 out of 5, across the three functional (4.7), technical (4.7), and clinical (4.3) dimensions.
Conclusion: This study provides a roadmap for the transition from predictive AI to agentic AI in hemato-oncology. The proposed multi-agent architecture, while preserving the human-in-the-loop principle, ensures reliability, explainability, and regulatory compliance, and can reduce the cognitive burden on tumor boards while democratizing access to precision medicine.
Methods: This conceptual study was conducted using the Design Science Research methodological framework. The proposed architecture was developed based on the convergence of three reference frameworks, including the ReAct reasoning paradigm, the HL7 FHIR R4 standard, and the NIST AI RMF 1.1 framework. The system architecture was described using Unified Modeling Language (UML) 2.5, and four specialized agents (Data Parser, Genomics, Guideline RAG, and Orchestrator) were specified based on a standard seven-component design pattern. The architecture was evaluated using the multidimensional RAMSES framework by an eight-member multidisciplinary expert panel.
Results: The proposed architecture was designed with five distinct layers (data acquisition, agent coordination, knowledge and tools, safety and governance, and presentation). The system employed a multilayered Defense-in-Depth security approach with four independent protective layers, which reduced the hallucination rate from 18.4% to 3.2%. An application scenario based on an AML case showed that the system was able to process a complex clinical file in 3 minutes and 21 seconds and produce a 2,847-word structured report with a confidence score of 0.82. The RAMSES evaluation yielded an overall average score of 4.4 out of 5, across the three functional (4.7), technical (4.7), and clinical (4.3) dimensions.
Conclusion: This study provides a roadmap for the transition from predictive AI to agentic AI in hemato-oncology. The proposed multi-agent architecture, while preserving the human-in-the-loop principle, ensures reliability, explainability, and regulatory compliance, and can reduce the cognitive burden on tumor boards while democratizing access to precision medicine.
Keywords: Artificial intelligence, Multi-agent system, Hemato-oncology, Tumor board, Clinical decision support, System architecture
: Original Article |
Subject:
AI in Medicine
Received: 2026/05/7 | Accepted: 2026/06/20 | Published: 2026/06/30
Received: 2026/05/7 | Accepted: 2026/06/20 | Published: 2026/06/30
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
- Iranian Journal of Blood and Cancer
- Journal Tel: +989369936116
- Publisher Tel: +982122275917
- Mail: Info@ijbc.ir
