Editor’s note: On January 18, 2021, the UK Parliamentary Office of Science and Technology released a research briefing titled “AI and Healthcare”, authored by John Smeaton and Lorna ·Christie, this briefing introduced the value and broad application prospects of artificial intelligence in the healthcare field, discussed the potential impact of artificial intelligence applications on the service quality, cost and practitioners of the medical industry, and targeted many aspects such as security and legal issues that artificial intelligence technology may be widely used in the healthcare industry were discussed. This article is an excerpt from the main content of the briefing.
Regarding the definition of artificial intelligence (AI) systems, there is currently no conclusive conclusion, which generally refers to intelligent systems that can perform tasks that require human intelligence to complete tasks. The artificial intelligence system is supported by algorithms, which can analyze a series of data and get results that meet the requirements. In recent years, due to the development of machine learning simulation algorithms, deep learning and other technologies and the improvement of the quality and volume of training data, the performance of artificial intelligence systems has been developed by leaps and bounds.
At present, academics, medical professionals and decision makers are full of expectations and interest in the application of artificial intelligence technology in the healthcare field. Research shows that artificial intelligence can improve the quality of medical services, reduce medical costs, and reduce the time consumption of employees on administrative tasks. Artificial intelligence technology has been applied in many aspects of the healthcare field, including assisted diagnosis, health monitoring, wearable devices, and smart medical care. The East Midlands Imaging Network (East Midlands Imaging Network) and its partners are testing an artificial intelligence tool that analyzes mammograms, which can screen for symptoms of breast cancer and perform various types of scanned images. Although there are already some artificial intelligence systems on the market, there is no large-scale use of artificial intelligence technology in the British National Health System (NHS). In the British National Medical System, most of the artificial intelligence products used in healthcare are still in the research or development stage, and some of these products are currently in the trial or evaluation stage.
Although artificial intelligence is not widely used in the British national medical system, the British government is full of confidence in the application prospects of artificial intelligence in the medical field. In the 2017 industrial strategy, the British government stated that it will use big data and artificial intelligence to achieve the goal of “transforming the way of disease prevention in the medical system” and achieve the prevention, early diagnosis and treatment of chronic diseases by 2030. In 2018, the UK invested 50 million pounds in 5 new national centers of excellence in medical and nursing to study how to use artificial intelligence to improve medical pathology diagnostic imaging; in the same year, the UK government issued a code of conduct for data-driven medical and nursing technology aiming at promoting the use of artificial intelligence and digital technology in the country’s medical system. In 2019, the British government established the National Healthcare System Digitalization Center (NHSX), which is responsible for formulating relevant policies and guiding the practice of promoting artificial intelligence technology in the healthcare field. It also spent £250 million to support the construction of artificial intelligence laboratories and the R&D and application of smart technology in the British National Medical System. In 2020, the British government will continue to allocate 50 million pounds to support the National Center for Medical and Nursing Excellence to conduct research on using artificial intelligence to better respond to the spread of the new coronavirus (COVID-19). In addition, the improvement and enhancement of artificial intelligence and digital technologies in the field of health care have been identified as the priorities of the UK’s National Medical System Digital Center in 2019.
In the medical field, there are a large number of medical images that require professional analysis. Through deep learning of structured data sets and standard image formats, artificial intelligence systems can be trained to recognize and interpret various types of medical images. The current applications of artificial intelligence technology in medical imaging include:
Radiology: The artificial intelligence system can be used not only to detect fractures and tumors in X-ray images, but also to describe and detect the results of CT scans of the patient’s head, so as to determine whether the patient has stroke, traumatic brain injury, dementia, etc.
Pathology and endoscopy: The artificial intelligence system can analyze the diseased tissue samples under the microscope to distinguish benign and malignant tumors. Not only that, the artificial intelligence system can also assist doctors in identifying cancerous tissues and precancerous polyps in real-time colonoscopy videos.
Ophthalmology: Through the diagnosis and monitoring of retina photos, artificial intelligence can identify retinopathy and age-related macular degeneration caused by diseases such as glaucoma and diabetes.
Daily Affairs Management
In daily work, medical staff spend a lot of time on administrative management and clinical matters. A survey conducted by the British National Health System (NHS) shows that artificial intelligence technology can help medical staff improve the efficiency of handling the daily management of the hospital and the level of office automation. For example: Voice recognition technology can transcribe medical records and medical history described by patients, and send reminders to patients who missed appointments. For complex logistics and warehousing issues, issues such as material management and time lists can also be handled using artificial intelligence technology to improve efficiency.
Treatment planning and patient monitoring
The decision support system is a software-based artificial intelligence tool that can provide data support and suggestions for clinicians to make medical decisions. The artificial intelligence system can assist doctors in prescribing, assisting in diagnosis and identifying the risks of various complications for patients. The decision support system originated from pre-programmed rules based on clinical knowledge and clinical operation guidelines in the 1970s. Currently, this pre-programmed rule has been widely used. The current research and development of decision support systems focuses on using the deep learning capabilities of machine learning to improve the system’s ability to learn patient data and clinical literature. In this way, the artificial intelligence system can directly monitor the health of patients. In major hospitals in the UK, the use of cameras and wearable sensors for early warning of symptoms such as pressure ulcers, confusion, and circulatory failure has been carried out; high-risk patients outside the hospital can also be monitored and observed through remote equipment to avoid accidents.
Other patient-oriented applications
Some voice assistants and text-based chat robots can directly help patients check their physical condition and get treatment. Mobile phone applications combined with wearable sensors and other devices can also help patients self-manage diseases such as respiratory system, diabetes or epilepsy. Artificial intelligence can be embedded in these systems to help doctors track the patient’s condition in real time and provide tailored healthcare guidance. Similar artificial intelligence systems can also help patients manage their own electrocardiograms (ECGs) and perform urine tests.
Impact on medical costs
Using artificial intelligence to automate daily affairs management and clinical tasks can reduce medical expenses and improve the efficiency of medical services. Although the medical costs that artificial intelligence can save vary in each link, according to a survey by the Public Policy Institute in 2018, artificial intelligence and automation can save the British National Health System (NHS) 12.5 billion pounds by shortening the working hours of employees.
Some studies have also reported that artificial intelligence systems can perform diagnostic tasks like skin cancer, diabetes, and retinopathy as well as clinicians. This means that the disease can be diagnosed earlier and more accurately, and the future treatment costs will be greatly reduced. However, some researchers are also concerned about the learning ability of artificial intelligence, because there are currently few studies to examine the performance of artificial intelligence systems in actual clinical environments. In order to solve this problem, new reporting standards for research and evaluation of clinical behavior have been introduced to better evaluate the performance of artificial intelligence under actual clinical conditions.
Impact on patients
The application of artificial intelligence technology can make the diagnosis of diseases earlier and more accurate, and patients can also receive treatment before complications occur, thereby improving the health of patients. There is also some evidence that the monitoring equipment and applications installed in the home can enable patients to participate more actively in formulating their own treatment plans, thereby allowing patients to become more active and active in medical treatment, and improve long-term self-management of health and cure diseases. However, some stakeholders worry that the application of artificial intelligence may make the medical system less human. Public opinion research shows that people believe that human empathy is an important part of medical care, but the application of artificial intelligence systems will not only harm the doctor-patient relationship, but also allow doctors and patients to have more opportunities and channels for communication and exchanges. Although some people think that doctors are more comprehensive and systematic than artificial intelligence in diagnosing and treating diseases, some stakeholders suggest that more automation and artificial intelligence equipment should be used when dealing with routine things, so that the staff have more time to spend on patients to improve the patient experience. And artificial intelligence can formulate and implement customized care plans for different patients’ conditions.
Impact on healthcare
In order to make better use of artificial intelligence technology, healthcare personnel need new knowledge and skills training. For example: Medical staff need to learn new knowledge and technology to operate an artificial intelligence system and understand the principles of its operation and the limitations of the system itself. Increasing the level of digitalization in the medical industry will help the development and popularization of artificial intelligence technology. Not only that, the medical industry also needs a large number of positions and roles proficient in data programming and information management. A forecast by PricewaterhouseCoopers (PwC) shows that as the application of artificial intelligence systems in the medical industry grows, the medical industry in the UK will increase 22% of employment. The British Health Education Agency (HEE), which specializes in the training of health care personnel, has launched a program dedicated to digital skills training for medical system leaders and clinicians. The plan includes the establishment of digital academies in the British national medical system and “The establishment of the Topol Digital Scholarship”. Institutions such as the Association of Clinical Informatics and the Federation of Informatics Professionals are working to develop professional training to improve the ability of medical staff to use information and digital technology.
Challenges to social ethics and laws caused by the application of artificial intelligence
In 2020, a ranking by consulting firm Oxford Insights showed that the UK’s preparations for artificial intelligence ranked second in the world, second only to the United States. However, some stakeholders emphasized that the British National Health System (NHS) has many difficulties in the long-term innovation and development, such as the lack of special funds and organizational fragmentation. Moreover, there are also some technical and ethical issues related to the application of artificial intelligence in the British National Health System (NHS).
Safety and effectiveness of artificial intelligence applied to the healthcare industry
While artificial intelligence systems have shown great potential in improving patient experience and improving the quality of medical services, we should also realize that if artificial intelligence systems have defects in design or improper use, they will also cause huge safety risks. When encountering unexpected situations that are not covered by the training data, the artificial intelligence system may give dangerous suggestions and instructions. For example, there are reports that some artificial intelligence robots have lost the function of simulating heart fibrillation and sexually abused children during testing. If the control parameters of a system are set to be too sensitive, it may lead to over-diagnosis of patients or even dangerous clinical interventions, and medical costs will also increase.
Perhaps the artificial intelligence system performs very well in the development and testing process, but there are still various problems and challenges to be solved in the application of medical practice, which undoubtedly hinders the promotion and application of artificial intelligence systems. For example: Google’s retinal artificial intelligence detection system performed well in the development process and even surpassed human professional ophthalmologists, but it performed disappointingly in the medical practice of several hospitals in Thailand. The main reason is that the quality of the retinal scan imaging results in actual operation is much worse than the scanned images during the training of the artificial intelligence system. Of course, there are other problems such as the interaction between humans and artificial intelligence systems. Certain professional biases of some medical staff may cause them to trust the automated decision-making of the system, but they may also distrust it.
The National Medical System Digital Center (NHSX) and the National Center for Medical and Nursing Excellence Institute have cooperated with other relevant institutions and issued evaluation standards for digital medical technology. These standards enumerate safety, medical efficiency, availability, and cost-effectiveness so that medical service providers can require developers to meet such conditions before purchasing artificial intelligence systems.
Privacy protection and data sharing
Since the use of patient data will be subject to many existing laws and regulations, the use of large amounts of data to develop artificial intelligence systems will cause a series of privacy issues. For example: In 2017, the Office of the Advisory Commissioner (ICO) found that Royal Free Hospital had failed to comply with the provisions of the data protection law and leaked identifiable patient data to the medical giant Deep Mind for kidney injury diagnosis System development. There is evidence that the public lacks awareness of incidents such as how medical institutions share patient data, and they are also skeptical about the sharing of patient data, especially those between medical institutions. A survey in 2018 showed that 50.3% of 2080 British adults are willing to share anonymous personal health data; 12.2% are willing to share personal health information with research that is used to improve medical services.
There are major differences in data sharing agreements and data formats between different institutions in the UK National Health System (NHS). Some stakeholders worry that the leaders of the NHS system lack the expertise to reach a consensus on the data sharing agreement, so that the value of patient data controlled by the NHS system cannot be fully realized. In 2020, the British government established the National Healthcare System Digitalization Center (NHSX) to improve the fluidity and sharing of data. The establishment of this organization is designed to ensure that the data sharing partnership within the medical system can be maintained, and the entire national medical service system can benefit from data sharing.
Artificial intelligence systems require massive and high-quality training data sets to produce accurate output. Inaccurate or incomplete data will cause the output results of the artificial intelligence system to seriously deviate from the expected results. Training data usually needs to be stored in a structured digital format so that the data can be more easily recognized by machine learning algorithms. However, the data quality and format of different organizations in the British National Medical System vary greatly, depending on the degree of electronic and standard data. For example: Although in 2017 54% of the UK National Health System (NHS) trust agencies reported that medical institution personnel can rely on digital technology to electronically record all the information they need, the secondary health care in the UK National Health System (NHS) The unit still uses a large number of paper-based methods for recording. In addition, many IT systems used by the British National Health System (NHS) cannot communicate and interoperate with other systems, which makes data collection more difficult. In this regard, the long-term development plan of the British National Medical System (NHS) has made the realization of system connectivity and interoperability a priority goal, so as to achieve data collection and sharing. Under this plan, it is expected that by 2024, all medical units in the British National Medical System (NHS) will achieve “digital information”.
Commentators expressed concern that the widespread use of artificial intelligence technology and other technologies in the healthcare industry may cause potential risks of cyber attacks on medical systems. During the development of artificial intelligence, the sharing of large data sets within the medical system and external developers will greatly increase the risk of data leakage. Hackers and other forces may manipulate the artificial intelligence system to tamper with the output and interfere with the medical system or carry out medical fraud and even steal personal data provided by patients during the research and development process.
Accountability Mechanism and Legal Responsibilities
The survey shows that the public has different attitudes towards artificial intelligence or other automated decision-making systems due to concerns that the application of artificial intelligence will lead to the ambiguity of medical responsibilities and that doctors in charge shirk their responsibilities to patients. In 2016, a survey conducted by PricewaterhouseCoopers (PWC) on 1,2003 adults in 12 countries showed that 39% of British survey respondents indicated that they are willing to use artificial intelligence systems for diagnosis or treatment and to obtain medical care advice; 50% Of British survey respondents expressed their reluctance to do so. At present, almost all artificial intelligence systems provide advice to clinicians in practical applications, and then doctors make final medical decisions based on their knowledge and experience. A medical science seminar involving 53 patients and members of the public stated that artificial intelligence should assist clinicians in making decisions, rather than leading medical decisions.
From a legal point of view, if a clinician causes harm to a patient by adopting the recommendations of an artificial intelligence system, then clinicians, system developers and medical service providers will all face criminal prosecutions and civil claims, and clinicians will also face medical associations However, there is still no precedent for how to deal with such incidents. At present, there is no professional regulatory agency to formulate guidelines or regulations to regulate and guide the use of artificial intelligence. Professional academic institutions such as the Royal College of Medicine are worried that the application of artificial intelligence is difficult to clarify legal responsibilities and accountability. With the application of “black box” technology, it is more difficult to clarify the responsibility of decision-making. Under the general application of artificial intelligence, the process of medical decision-making will become very complicated, which makes it difficult for people to fully understand the process of a medical decision. Some stakeholders said that the future society needs a new legal system for artificial intelligence to guide and regulate the application and development of artificial intelligence.
How to develop and use artificial intelligence systems determines whether medical inequities will increase or decrease in the future. The artificial intelligence system can diagnose different symptoms based on the latest medical data and implementation standards and provide undifferentiated and standardized medical care recommendations, thereby reducing the difference in service quality between different medical institutions and improving patient experience. However, the artificial intelligence system still has algorithmic flaws and will provide suggestions that conflict with the data set. This situation may be caused by the total number of samples used by the system during the research and development period that cannot cover all real-world training data. For example: a common skin cancer research database mainly contains skin imaging data of white patients, but lacks skin imaging data of other skin color patients. Experts believe that the use of these data for training by machine learning systems will cause the system to fail to diagnose patients with black skin. The “Data Protection Law” requires the use of personal data to avoid discrimination in the process of research and development and treatment, and the “Equality Act of 2010” also prohibits discrimination and discrimination between different groups based on inherent characteristics.
Regulatory Issues of Artificial Intelligence
Artificial intelligence systems that are directly used for medical purposes will be included in the management of medical devices, in vitro diagnostic devices or removable implantable devices. In the UK, such items are regulated by the Medicines and Healthcare Products Regulatory Agency (MHRA). According to the 2018 Brexit Act, relevant EU laws will be retained during the transition period of Brexit, during which the “EU Device Management Regulations” will continue to take effect. In addition to the above-mentioned regulations, the British government also issued regulatory guidelines for medical devices circulating in the domestic market in January 2021. In addition to Northern Ireland, relevant laws in the UK will be formulated in accordance with the Pharmaceuticals and Medical Devices Act 2019-2021. The British government stated that the new regulations will focus on the safety of medical devices and drugs and better promote the application and promotion of new technologies including artificial intelligence in the medical industry.
The use of personal data in the field of artificial intelligence will be supervised by the Office of the Information Commissioner (ICO). However, the use of data within the British National Health System (NHS) has stricter security standards, and the research and development of artificial intelligence will be recognized For medical research and development, permission from the medical research and development management department is required. The British Medical Quality Council has stated that any artificial intelligence system that diagnoses and treats patients without human intervention needs to be registered. Relevant persons in the medical industry believe that the existing British medical R&D regulatory system involves many departments and the process is cumbersome, which is difficult to promote the development of medical innovation and has become an obstacle to innovation. The Artificial Intelligence Laboratory of the British National Health System (NHS) is funding projects to simplify regulatory procedures, including the creation of a multi-agency consulting service system, hoping to provide a more convenient one-stop contact channel for AI developers seeking guidance .
Under the existing regulatory and legal systems in the United Kingdom, the development and application of machine learning algorithm systems are extremely difficult. Although the system can continue to learn and optimize algorithms and output results when new data is obtained, it wants to ensure that the system is safe and efficient. There are still many difficulties in its operation. The US Food and Drug Administration has issued regulations that allow R&D personnel to pre-design a series of safety procedures for future AI system changes, but this is not enough in the UK. As a partner of the United States, the National Standards Association has considered how to formulate international medical device management standards to deal with and solve various situations and challenges that may be faced by the development of artificial intelligence in the United Kingdom.