With technology developing rapidly and databases exploding with massive clinical data, many medical device manufacturers have been searching for different approaches to enhance the healthcare system with well-known experiences. More and more manufacturers are taking advantage of this new artificial intelligence technique and training the software to give diagnostic information. This kind of software is identified as Artificial Intelligent/ Machine Learning-Based Software as a Medical Device, or so-called AL/ML-Based SaMD.

In September 2020, Taiwan FDA announced a Product Registration Guidance for Artificial Intelligent/ Machine Learning-Based Software as a Medical Device. In the guidance, it illustrates the concepts and algorithm structures that manufacturers shall consider when developing the Al/ML-Based SaMD. It also applies to medical devices that uses AI or ML-Based techniques as part of the function. It is not intended to define the medical device categories or risk levels.

The below summarizes some of the main points and requirements given.

  • 1. Defining the software as:
  • - Artificial Intelligence, AI
  • - Machine Learning, ML
  • - Deep Learning
  • 2. Description of software functions:

Functional types, including-

  • - Computer Assisted Detection, CADe
  • - Computer Aided Diagnosis, CADx
  • - Computer Aided Triage

Algorithm framework-

  • - Design
  • - Training method
  • - Testing principle and algorithm framework
  • 3 Data Limitation
  • - Shall include the training method, structure, and process description.
  • - Shall describe the training module database, in population, clinical meaning, output type, output method, and other additional clinical significances
  • 4. User Environment and Data Management Safety


  • 5. Functional Verification and Validation

Shall refer to “Software Medical Device Guidelines”, including-

  • 1. Level of concern
  • 2. Software description
  • 3. Device hazard analysis
  • 4 .Software requirements specifications
  • 5. Architecture design chart
  • 6. Software design specification
  • 7. Traceability analysis
  • 8. Software development environment
  • 9. Verification and validation documentation
  • 10. Revision level history
  • 11. Unresolved anomalies/ bugs or defects
  • 6. Clinical Significance

Scientific evidence shall be provided to establish the applicability and suitability of the software specifications. It is recommended to include the below concept in the study protocol-

  • 1. Intended use
  • 2. Study objectives
  • 3. Patient population, e.g. age, ethnicity, race…
  • 4. Number of clinicians and qualifications
  • 5. Description of the methodology used in gathering clinical information
  • 6. Description of the statistical methods used to analyze the data
  • 7. Study results



TFDA Notification #1091607253 Product Registration Guidance for Artificial Intelligent/Machine Learning-based Software Medical Device