Research
Medical research and data science
Promote research in Medical AI, smart healthcare, and medical data science; establish collaboration across institutions, shared specifications, and traceable evaluation methods.
Medical AI: beyond Clinical AI
Clinical AI is an important area of Medical AI, but it does not encompass the Society’s entire scope. The following directions for work (DRAFT) are based on the charter discussion draft and the new scope; they do not indicate that the relevant services or research have begun.
Care & Health
Clinical care and health promotion
Explore appropriate applications in care processes, human–AI collaboration, and health promotion; clinical decision support must still meet the requirements for its respective intended use, evidence, and professional oversight.
Quality & Operations
Quality, patient safety, and healthcare management
Evaluate the value of AI in terms of quality improvement, patient safety, workflow burden, and healthcare management needs, also including administrative and digital workflows outside diagnosis and treatment.
Education
Education and interdisciplinary talent
Enable physicians, healthcare professionals, researchers, engineers, and students to learn together and build the capacity to collaborate through courses, workshops, and academic exchange.
Governance
Data governance and public trust
Incorporate privacy, information security, ethics, fairness, human oversight, and auditing into research and applications; clearly explain model limitations and the allocation of responsibility.
Translation
Translation of innovation into practice, and public engagement
Promote industry–academia collaboration, evaluation, and dissemination of results while maintaining research integrity, conflict of interest management, and academic independence.
Existing clinical engineering blueprint: three North Star goals
The clinical engineering goals of the v2.0 blueprint are retained below as one implementation pathway for Medical AI work; they do not represent the Society’s entire scope or completed results. The schedule and resource allocation for the expanded scope remain pending review and approval.
Within one year
Develop an AI workflow that makes a noticeable difference to physicians
Begin with documentation-burden reduction and rule-based CDSS: pre-visit SOAP drafts, pre-clinic summaries, and checks for heart failure GDMT and atrial fibrillation anticoagulation.
Within three years
Develop a clinical AI platform that can undergo multicenter validation and seek regulatory authorization
Federated multicenter validation with data remaining within hospitals, and prospective silent trials, to support the multicenter evidence needed for SaMD regulatory authorization.
Within five years
Become a center for Medical AI standards and evidence in Taiwan
A Taiwan-specific Model Card standard, a CQL rule library as a public resource, a national clinical AI deployment registry, and process certification.
Five pillars
The Society’s work is organized around five pillars. See Five pillars for details.
AI education and interdisciplinary talent development
Plan education in AI literacy, practical skills, and governance for healthcare, academic, research, engineering, and management professionals; the existing clinical course pathway, expanded curriculum, and certification policy remain pending review and approval.
Medical AI research and trustworthy evidence
Focus on fit-for-purpose evaluation and cross-institutional research; the existing clinical engineering prototypes explore federated validation and prospective silent trials and do not yet constitute actual multicenter research results.
Healthcare applications, quality, and process improvement
Across care, health promotion, and healthcare management, first clarify the problem and measurable value; the existing clinical workflow and CDSS are among the demonstration areas within this scope.
Governance, policy, and translation of innovation into practice
Support collaboration through ethics, conflict of interest management, and academic independence; for applications involving medical devices, assess the regulatory pathway separately according to the intended use. Not all Medical AI should be regarded as SaMD.
Secure digital infrastructure and Agent/MCP
Forward-looking exploration under strict controls: use an internal checklist, static analysis, and testing with synthetic data to build prototypes of an MCP repository and a permission-tiered agent runtime.
Five-layer architecture at a glance
The existing clinical engineering pathway comprises 20 projects (P-01–P-20) organized into five layers. This is one starting point for Medical AI; the name change does not alter any project’s intended use, maturity, or evidence status.
ACore systems layer
PHI scrubbing gateway · FHIR minimization · Audit ledger · Data dictionary · Evaluation framework (P-01–P-05)
BClinical workflow layer
Pre-visit SOAP · Clinic summaries · Heart failure GDMT · AF anticoagulation · CAD/ACS secondary prevention (P-06–P-10)
CMulticenter evidence layer
CQL rule library · Federated validation · Prospective silent trials · Model Card registry (P-11–P-14)
DAgent/MCP layer
Medical MCP/skill repository · Clinical agent execution framework (P-15–P-16)
ERegulatory authorization and governance layer
SaMD document generator · Post-market surveillance · Institutional review board (IRB)/governance documents · Society operations platform (P-17–P-20)
View P-01–P-20 projects and evidence → (Mandarin only) Interpret maturity and evidence → (Mandarin only)
Five non-negotiable governance rules
Governance principles shared across the platform; refer to the evidence inventory for the extent of implementation and limitations of each v0.1 prototype:
- Data must remain within the hospital — The target architecture prohibits patient-level data and raw PHI from leaving the hospital environment.
- Pseudonym-to-identity mapping tables (re-identification keys) must remain within the hospital — The target architecture requires these tables to be stored separately in a network-isolated environment; they must not be sent out with the payload.
- Keep an audit record of every call — P-03 already has a tamper-evident ledger prototype; production integration for each workflow remains to be completed.
- Physician-in-the-loop — Physician confirmation state machines have been tested in multiple prototype implementations; this does not yet indicate deployment in an actual EMR workflow.
- Retain the physician’s override reason — This is a mandatory design requirement for, and a quality-monitoring input to, all future production workflows.
View governance and responsibility boundaries (Mandarin only) · Safeguards for patients / the public (Mandarin only)
Recent updates
Name and direction | 2026-09-06
Name confirmed: TSMAI
Taiwan Society of Medical Artificial Intelligence (TSMAI). The Society’s scope has expanded to Medical AI; the charter and organizational structure remain discussion drafts. The name change does not indicate completion of registration as a legal entity or completion of the various services.
Engineering
P-01–P-20 v0.1 engineering baselines
Initial prototypes and test code have now been completed for all projects (P-01–P-20). Please refer to the documentation on each project’s page for its technical maturity, clinical application contexts, and regulatory progress. These prototypes are for technical demonstration only; they have not yet obtained permission for clinical deployment or passed regulatory review.
Education
Level 1 course — in planning
An AI literacy course is in preparation, using the Ministry of Health and Welfare’s official guidelines, Guidelines for the Application of Generative Artificial Intelligence in Medical Institutions (unofficial translation), as the basis for course materials.
Full news and events page — in preparation