Continuous learning AI-MDs will also need to submit the following:
- Description of the continuous learning process during deployment.
- Safety mechanisms meant to detect anomalies and any inconsistencies that would affect the products performance.
- Comparison between the real world data that will be collected and the original training set to ensure that ongoing evaluation is consistent with the original data evaluated.
- Processes to ensure data integrity, reliability and validity.
- Software version controls in case roll-backs to a previous version is required.
- Validation and verification activities to ensure performance remains within pre-defined boundaries.
- Additional traceability requirements dependent on deployment structure, clinical use or user training requirements for different versions.
CLINICAL EVIDENCE REQUIREMENTS
The level of scrutiny and clinical data required during the application process will be determined by the significance of the information being produced. Decision making devices, locally referred to as a “sole determinant” will require more evidence, like clinical studies or data, than those in more of a supporting role. Support devices are broken down into two categories; informing or driving. Those that “inform” use early detection mechanisms that are then confirmed by a practitioner and/or additional tests. Most common types of informing devices are wearable technologies that can detect blood pressure, heart arrhythmias or sleep patterns. “Drive” devices, on the other hand, typically assist practitioners to mitigate or prevent diseases or conditions, such as software programs that help oncologists evaluate mammograms.
IMPORTING AND INVOICING FOR MEDICAL DEVICE SOFTWARE
It is important to note that software as a medical device (SaMD) even without physical forms must still adhere to standard medical device labeling, importing and invoicing requirements. For example, the manufacturer will still need to list a licensed Medical Device Importer on the Singapore Medical Device Register (SMRD) listing. That entity will have primary post market vigilance responsibilities in Singapore. Furthermore, medical device sales must be invoiced by a local entity with a Medical Device Sales License that will be responsible for VAT payments. These requirements apply to software that is downloaded directly from the internet.
AI IN THE MEDICAL DEVICE/IVD INDUSTRY
It is important to understand the regulatory requirements for software registration as Software as a Medical Device (SaMD), AI and Machine Learning (ML) become increasingly integrated into medical devices and health care. These technologies can improve the accuracy of diagnoses, optimize patient experience in hospital systems, perform mass data collection and enhance robot-assisted surgeries by using adaptive software that learns to recognize new patterns.
The evolving nature of these technologies has prompted regulatory bodies to create new registration guidelines that apply for the duration of the software’s life cycle. These guidelines will likely adapt as the uses of AI expand in the coming years. Singapore is the first Asian Market to provide guidelines for AI, but other countries will need to release guidance because current regulatory controls are not sufficient for this robust technology. Asia Actual will report on future guidance as it is released.