The Taiwan Food and Drug Administration (FDA) has published technical guidelines for medical device software inspection and registration of Artificial Intelligence and Machine Learning technology. The new approach is based on the regulations adopted in the United States, Japan, South Korea, and other countries, as well as on the guidance documents developed by the International Medical Device Regulators Forum (IMDRF). The document is intended to provide medical device manufacturers involved in the development of medical software products with important information related to the evaluation of the software and applying for the artificial intelligence/machine learning medical software registration in Taiwan.
As it is stated in the present guidance, the medical device software based on the artificial intelligence/machine learning technologies should comply with the applicable regulations including the Pharmaceutical Affairs Acts. In particular, the guidance describes the approach to the product description, safety and performance evaluation, and other important aspects associated with the medical device software. The authority also mentions that since the whole software sphere in general and medical software, in particular, develops swiftly nowadays, the present regulatory approach described herein could be subject to certain changes reasonably necessary to ensure the safety and effectiveness of the medical software allowed to be marketed and used in the country. The Taiwan FDA also reserves a right to request the medical device manufacturers to provide additional information related to the medical software in question, as well as the safety and performance verification and evaluation data even if the requested information actually falls outside the scope of the present guidance.
The scope of the guidance covers the Artificial Intelligence / Machine Learning-Based Software as a Medical Device (AI/ML-Based SaMD) which uses clinical data (assay data, databases, film, or images) as sources, and through the artificial design of the software learning mode or training method makes the program model. The scope of the document also covers the medical software that learns independently to adapt its performance, as well as all the cases when the AI/ML technologies are used to provide some functions of the medical device even not being a separate medical device. The guidance describes the examination procedures and registration of the AI/ML-Based SaMD. However, all matters associated with the classification of medical software are covered by another document providing the guidelines for classification of medical software issued previously by the Taiwan FDA.
AI/ML-Based SaMD: Key Definitions
First of all, the present Taiwan FDA guidance outlines the definitions of the most important terms and concepts associated with medical software registration in Taiwan, namely:
- Artificial Intelligence, AI – a technology allowing a machine or computer program to simulate a human from the appearance of intelligent behavior, including voice conversation, visual recognition, motion control, ability to learn, make decisions, and also the self-correction.
- Machine Learning, ML – a technology allowing the computer (software) to avoid the excessive processes by using designing algorithms and data training instead. ML technology allows the computer to learn independently using the data, and to improve the calculations through the training experience. It also includes the methods used to imitate various calculation methods of human learning function, e.g. regression analysis, decision trees, and neural networks.
- Deep Learning – a branch of machine learning that uses neural network structure (e.g. multi-layer network, or convolutional neural network) and a large volume of training data allowing the network to achieve significant results.
Basic Concepts of AI/ML-Based Medical Software
When applying for the registration of AI/ML-Based medical software registration in Taiwan, the manufacturer (developer) shall provide the full set of documents covering such aspects as the software functional description, software architecture, software adoption evolution (adaptive or locked algorithm design). Besides the technical details mentioned hereabove, the submission should contain:
- the product’s intended use,
- the indications for use,
- contradictions and limitations,
- functional value (e.g. detection rate, false-positive rate, false-negative rate, the time required for testing).
According to the guidance, the aforementioned technologies could be used for the following purposes:
- Computer-Assisted Detection (CADe) – the system uses artificial intelligence/machine learning to analyze medical images and data to detect abnormal values,
- Computer-Aided Diagnosis (CADx) – the system uses artificial intelligence/machine learning technologies to process data and provide diagnostic options and risks,
- Computer-Aided Triage – the system performs rapid screening to reduce the workload of the clinical staff.
Software-Specific Requirements for Medical Software Registration in Taiwan
As it was already mentioned before, the medical device manufacturer applying for the registration of the AI/ML-Based SaMD shall provide the exhaustive information about the software product subject to review. The scope of information provided should be sufficient for the regulating authority to evaluate the risks and clinical benefits associated with the particular medical software. Due to the complexity of the systems, the Taiwan FDA emphasizes the importance of technical documentation to be provided. In particular, the authority requires to provide a detailed description of the detection principles and structure of the algorithm. In the case of «black box» systems, the manufacturer shall provide the details regarding the design and training of the algorithm.
The medical device manufacturer applying for the registration of the AI/ML-based medical software in Taiwan shall supplement its initial submission with the following:
- The description of the original network architecture of the medical software and its overview.
- Information about the training methods, including the indication of the intended use, basic data sources, learning methods, as well as the approaches used to verify the data and evaluate the performance and effectiveness depending on the data used.
- The description of the algorithm architecture.
- The indications on special data restrictions (due to the nature of the AI/ML technologies, the quality of input data could impact dramatically the quality of the final results, their accuracy, and reliability). In particular, the manufacturer shall indicate the methods to be used to collect the data in order to ensure the correct operations of the medical software. The most important aspects to be covered in this context are related to training, verification, and testing.
- Use environment and information security – the manufacturer shall duly implement the appropriate cybersecurity measures necessary to protect the software against any unauthorized third-party intervention, including when various wireless connection technologies are being used.
- Functional verification – software validation report demonstrating that the AI/ML-based medical software operates as intended. This part should also cover such aspects as the risk level associated with the device (level of concern), the description of the software, device hazard analysis, software requirements and specifications, architecture design chart and software design specification, traceability analysis, software development environment, revision history and bugs or defects.
The guidance also describes the requirements applicable to the output results the medical software provides. According to the document, the output results should match the intended use of the software. They should also be presented in the appropriate forum.
Summarizing the information provided here above, the guidance on the AI/ML-based medical software issued by the Taiwan FDA outlines the most important aspects associated with the medical software products based on the novel technologies, including the documents to be provided by the medical device manufacturer (developer) when applying for the medical software registration in Taiwan.
How Can RegDesk Help?
RegDesk is a next-generation web-based software for medical device and IVD companies. Our cutting-edge platform uses machine learning to provide regulatory intelligence, application preparation, submission, and approvals management globally. Our clients also have access to our network of over 4000 compliance experts worldwide to obtain verification on critical questions. Applications that normally take 6 months to prepare can now be prepared within 6 days using RegDesk Dash(TM). Global expansion has never been this simple.
Want to know more about our solutions? Speak to a RegDesk Expert today!