Principles

• Applies to fully developed medical-AI systems intended to interface with CMUH HIS and collect clinical data for performance evaluation.
• Medical-AI developed in-house or holding a Taiwan FDA (TFDA) marketing authorization is given priority for trial.
• Medical-AI without a license must first obtain CMUH Institutional Review Board (IRB) approval before applying for trial and subsequent production.
• High‑risk medical-AI (Class III or IV) must obtain IRB approval and complete clinical trials before it may apply for trial and production.

  Risk Class		Use Context for Inference Output
  		Treatment / Diagnosis	Drives Clinical Management	Informs Clinical Management
  Clinical Condition	Critical	IV	III	II
  	Serious	III	II	I
  	Non‑serious	II	I	I
  Treatment / Diagnosis: Direct basis for a physician’s treatment or diagnostic decision. Drives clinical management: Triggers secondary assessment or intervention by healthcare staff. Informs clinical management: Reference only; final diagnosis/decision remains with healthcare staff.

  * Adapted from U.S. FDA risk‑stratification guidance.
• Applicants must:
  • Provide metrics and reference standards for the safety and effectiveness of the medical-AI;
  • Cooperate in continuous monitoring and periodic reporting to enable life‑cycle management;
  • Submit the Nine‑Point Transparency Disclosure and explainability analysis, publicly accessible to all stakeholders.
• The inference server may be provided by CMUH or by the applicant. When using a CMUH server, installation is the applicant’s responsibility.
• The medical‑AI system should be loosely coupled with existing hospital IT systems and must not disrupt their operation.
  • CMUH has issued the AI‑on‑HIS FHIR Implementation Guide for integration.
  • Alternative integration methods must first be reviewed for technical feasibility by the IT Office, and any additional costs borne by the applicant.
• The medical‑AI system must comply with the Cybersecurity Management Act, the Personal Data Protection Act, and relevant CMUH regulations and ISO frameworks.
• By default, medical‑AI must not connect to external networks.
  • If external connectivity is required, the applicant must submit a detailed security plan and bear associated costs.
  • Unauthorized external connections are strictly prohibited and will be reported to the Personnel Evaluation Committee.
• Trial and production applications should be submitted by CMUH employees. External organizations should apply via a CMUH partner department or contact the Center for assistance.

Internal Department Workflow

External Organization Workflow

Trial Application

A trial is a small‑scale evaluation of medical‑AI in the HIS. Clinical decisions remain solely at the discretion of healthcare professionals.

Steps

1. Download and complete the Medical‑AI Software Trial Application Form.
   • Fill out the Information‑System Security Baseline Self‑Assessment according to the required security level.
   • We recommend pre‑filling the Nine‑Point Transparency Disclosure Form.
2. Submit an “AI2 Medical‑AI Software Go‑Live Request” in the BPM e‑sign system and select Trial.
3. The Smart Healthcare Committee reviews:
   • Feasibility and necessity;
   • Integration technical suitability;
   • Ethics and regulatory compliance;
   • Safety, including cybersecurity and personal‑data protection.
4. Only after approval may the AI connect to HIS and start the trial.
5. System go‑live follows existing ISO procedures of the IT Office.
6. At trial end, evaluate performance per the plan in the application.

Flowchart

Production Application

Production refers to a trial‑approved medical‑AI whose performance meets requirements and is fully integrated into clinical workflows.

Steps

1. Download and complete the Medical‑AI Software Production Application Form.
   • Attach trial performance‑evaluation results.
   • Fill out the Information‑System Security Baseline Self‑Assessment.
   • Submit the mandatory Nine‑Point Transparency Disclosure Form.
2. Submit an “AI2 Medical‑AI Software Go‑Live Request” in BPM and select Production.
3. The Committee reviews and grants a production period.
4. Only after approval may the AI formally connect to HIS.
5. Go‑live follows standard ISO procedures.

Flowchart

Life‑cycle Monitoring & Management

• Evaluate performance periodically as defined in the production application.
• Before the production period expires, submit evaluation results to apply for extension.
• The Committee conducts periodic reviews and decides on continuation or retirement.
• Retired AI may reapply for trial after being retrained or adjusted.

Flowchart

AI2 Medical‑AI Software Go‑Live Request (Form Screenshot)