HONG KONG – With an aging population and a shortage of doctors, Japan is now working to develop artificial intelligence (AI)-based medicine faster than any other country in Asia.

Japan's lawmakers and regulators are paying particular attention to the changing nature of AI-based devices and systems. AIs learn over time, in large part by reviewing huge datasets. In theory, the larger the datasets the better the performance of the AI. This means that the performance of AI devices may change over time and these changes create unique challenges for regulators.

Japan, which has a very well-developed health care system and health care regulations, is rapidly moving forward with the adoption of AI. Japan has plans to invest more than $100 million to open 10 AI-based hospitals by the end of 2022. These model hospitals will use AI in all tasks, from updating patient records to assisting with diagnosis. The advanced technology will also be used to analyze blood tests, vitals, electrocardiography, DNA parsing and imaging.

Japan is already using AI in surgery.

Earlier this year, Tokyo Women's Medical University and Waseda University Joint Institute for Advanced Biomedical Sciences used AI in a brain surgery to treat essential tremor, a nerve disorder. The surgery team used Smart Cyber Operating Theater (SCOT), which blends AI and robotics tools. The AI advised on surgical options such as where the incisions should be made and provided doctors with a data stream from the medical tools.

While accelerating the use of the advanced technology, Japan has set up regulations to govern the development of AI-powered medical devices, especially focusing on device performance evaluation. As they accumulate data and deep learning, AIs adapt their performance. As the performance changes, current regulation requires a whole new set of government approvals. This can be a time-consuming process.

To push for a more effective approval process, the Japanese government created a committee and a consortium to develop quality control standards and evaluate the efficiency of AI devices.

In 2017, the government's Pharmaceuticals and Medical Devices Agency (PMDA) put in place a subcommittee of the Science Board on AI and its Applications in Medical Field. The subcommittee held six meetings and published a final report in 2018 in Advanced Biomedical Engineering. The report stated AI-based medical and health care devices and systems have unique characteristics in the viewpoint of regulatory science.

The characteristics include plasticity that causes changes in system performance by learning. This constant change creates a need for new regulatory concepts that take into account the timing of learning and the assignment of responsibilities to manage risks. A regulatory framework should also account for the unpredictability of system behavior in response to unknown inputs as well as the need for assurance of the characteristics of datasets used for learning and evaluation.

"As AI-based medical devices have some unique characteristics, we organized the subcommittee to support policies for the devices. Even [though] the subcommittee meetings ended in 2017, we will supplement the policies based on our meeting results," a spokesman at PMDA told BioWorld MedTech.

Data quality is paramount

Mamoru Mitsuishi, executive director and vice president of the subcommittee and professor at the University of Tokyo, said the biggest challenge is to assess the quality of the data collected for the AI to learn after the government approval.

"In general, the data used to measure the performance of AI-based medical devices before approval is collected by selected people, and the quality of the data is controlled. After approval, however, data is collected to update the performance of the AI-based medical devices," he told BioWorld MedTech. "Such data will be collected by users including non-specialists. Therefore, the data may include missing data and errors. Also, there may be data collected under situations that the device developers have not expected in advance."

For quality control after approval, the Ministry of Health, Labor and Welfare (MHLW) created a consortium to promote AI in the medical field. Last May, the consortium put forth an assessment index for medical imaging diagnosis supporting systems based on AI.

The standards require manufacturers to control the quality of devices while clinical sites are expected to maintain quality-control systems equal to those of the manufacturer and carry out countermeasures when the device performance falls below its lowest limits.

If the performance of an AI device changes, the manufacturer is expected to notify users. When the performance changes after an AI device is launched, the manufacturer is also expected to assure the quality of the device in regards to clinical and statistical issues.

The MHLW also introduced last May a revised bill to screen systems and accelerate the re-approval of AI-based medical devices. The revised bill requires developers of AI-powered device to provide details about how the device will be updated using new data and by managing and improving machine learning capabilities.

For reauthorization, the government will check if the machine performs up to specifications after any updates.

The bill remains under discussion at the National Diet, Japan's legislature, even though the original goal was to pass the bill during the regular session last June. A key goal is to shorten the authorization process for AI devices. The current process takes nine to 12 months.