Dwarka Sector 9, Institutional Area, Near Metro Pillar No.1160
Nearest Metro Station Dwarka Sector 10, New Delhi-110075
Phone: 011- 45636921/22/23
Website: www.tips.edu.in
Email: tips@tips.edu.in
ABSTRACT WHAT IS AI?
Artificial intelligence (AI) aims to mimic There is no universally agreed definition of
human cognitive functions. It is bringing a AI. The term broadly refers to computing paradigm shift to healthcare, powered by technologies that resemble processes increasing availability of healthcare data associated with human intelligence, such as and rapid progress of analytics techniques. reasoning, learning and adaptation, sensory We survey the current status of AI understanding, and interaction.1 Currently, applications in healthcare and discuss its most applications of AI are narrow, in that future. AI can be applied to various types of they are only able to carry out specific tasks healthcare data (structured and or solve pre-defined problems. unstructured). Popular AI techniques include machine learning methods for OVERVEIW OF THE MEDICAL structured data, such as the classical support ARTIFICIAL INTELLIGENCE (AI) vector machine and neural network, and the RESEARCH modern deep learning, as well as natural language processing for unstructured data. Recently AI techniques have sent vast Major disease areas that use AI tools waves across healthcare, even fuelling an include cancer, neurology and cardiology. active discussion of whether AI doctors will We then review in more details the AI eventually replace human physicians in the applications in stroke, in the three major future. We believe that human physicians areas of early detection and diagnosis, will not be replaced by machines in the treatment, as well as outcome prediction foreseeable future, but AI can definitely and prognosis evaluation. assist physicians to make better clinical decisions or even replace human judgement in certain functional areas of healthcare (eg, radiology). Guided by relevant clinical questions, powerful AI techniques can unlock clinically relevant information HEALTHCARE DATA hidden in the massive amount of data, which in turn can assist clinical decision Before AI systems can be deployed in making. healthcare applications, they need to be ‘trained’ through data that are generated In this article, we survey the current status from clinical activities, such as screening, of AI in healthcare, as well as discuss its diagnosis, treatment assignment and so on, future. We first briefly review four relevant so that they can learn similar groups of aspects from medical investigators’ subjects, associations between subject perspectives: features and outcomes of interest. These 1. motivations of applying AI in clinical data often exist in but not limited to healthcare the form of demographics, medical notes, electronic recordings from medical devices, 2. data types that have be analysed by physical examinations and clinical AI systems laboratory and images. 3. mechanisms that enable AI systems Specifically, in the diagnosis stage, a to generate clinical meaningful substantial proportion of the AI literature results analyses data from diagnosis imaging, 4. disease types that the AI genetic testing and electrodiagnosis (figure communities are currently tackling. 1)
MOTIVATION
The advantages of AI have been
extensively discussed in the medical literature. AI can use sophisticated algorithms to ‘learn’ features from a large volume of healthcare data, and then use the obtained insights to assist clinical practice. It can also be equipped with learning and self-correcting abilities to improve its accuracy based on feedback. An AI system can assist physicians by providing up-to- date medical information from journals, textbooks and clinical practices to inform proper patient care. In addition, an AI Figure 1 system can help to reduce diagnostic and The data types considered in the therapeutic errors that are inevitable in the artificial intelligence artificial (AI) human clinical practice Moreover, an AI literature. The comparison is obtained system extracts useful information from a through searching the diagnosis large patient population to assist making techniques in the AI literature on the real-time inferences for health risk alert and PubMed database. health outcome prediction. IMPACT ON HEALTHCARE
More engaged and discerning consumers
are exerting greater influence on health systems and driving new business models. Widening choices in a growing global This trend is opening the door for new health market have opened the doors for entrants from industries such as retail, new players like telecommunications to telecommunications, technology, and access the broad healthcare consumer base. wellness and fitness. At the same time, new products, services and delivery systems are helping to democratise and decentralise healthcare.
An era of patient-centric health systems is
emerging. We are seeing a general shift away from fragmented care to integrated models: organisations, communities and social care providers coordinating their services, with patients as active partners in their health across the continuum. PATIENT AND CONSUMER-FACING APPLICATIONS Informed consumers will demand increasing accountability, integrity and transparency from their health systems. Our analysis of mHealth reveals mobile is Meanwhile, the pressure is on governments positioned to have a huge impact on how to provide sustainable care in the face of healthcare is delivered. It offers anticipated significant increases in opportunities to address one of the most healthcare costs. The private sector offers pressing global challenges: making partnership opportunities to satisfying healthcare more accessible, faster, better consumer demands. As financial and cheaper. performance is tied to clinical outcomes, patient engagement will also become Mobile is accelerating trends in increasingly important. The private sector healthcare has the resources and innovative Three major trends already happening in technologies to meet the outcomes that healthcare lend themselves to the revolution informed consumers increasingly value. in mobile technology: These companies can invest in insights and have the versatility to adapt to satisfy consumer needs. • Ageing population: Ageing populations and chronic illness are driving regulatory reform. Public sector healthcare is seeking mHealth services are already replacing better access and quality, and it’s looking to visits to nurses and doctors. Wireless the private sector for innovation and technology will improve access to health efficiency. mHealth improves access and for people around the world. As the quality, and offers dramatic innovation and technologies improve and patients discover cost reduction. they hold the power of self-care on their personal devices, they will expect to have • Foundations already in place: The greater access to these services. Mobile foundations of industrialisation of technologies will improve the quality of healthcare are already in place — electronic care they receive while easing some of the medical records, remote monitoring and pressures on the health system. communications. ‘Care anywhere’ is already emerging. The platform for now be done through a mobile phone and mHealth is set. wirelessly sent to a physician.
• Personalisation: Healthcare, like other
industries, is getting personal. mHealth can Patients are taking a greater interest in their offer personal toolkits for predictive, care and are more willing to self-manage. participatory and preventative care. They want to be more empowered when it comes to their health. Leveraging available health information, new technology, and mobile health (mHealth), the empowered consumer knows more, wants more and is CURRENT MARKET SITUATION able to do more for themselves. An era of patient-centric health systems is emerging, with a shift away from More demanding and discerning consumers fragmented care to integrated models. are opening doors for new entrants in healthcare provision. Consumers are taking advantage of unprecedented access to information to become more diligent and informed about their health. The growing power of the patient as discerning consumer is creating new global markets and informing new models for care.
Patients are demanding more sophisticated,
convenient, transparent, affordable and ETHICAL AND SOCIAL ISSUES personalised service. As a result, an agile private sector has gained a strong foothold Many ethical and social issues raised by AI in the delivery and financing of healthcare. overlap with those raised by data use; In a recent PwC consumer survey, almost automation; the reliance on technologies half of respondents said they would more broadly; and issues that arise with the consider having procedures like wound use of assistive technologies and treatment, stitches or staples removed at a ‘telehealth’. retail clinic or pharmacy. One-third said they would get an MRI at a retail venue. • RELIABILITY AND SAFETY The underlying message is that patients are Reliability and safety are key issues where accepting greater accountability for their AI is used to control equipment, deliver health. treatment, or make decisions in healthcare. AI could make errors and, if an error is difficult to detect or Patients are also welcoming the flexibility has knock-on effects, this could have that technology brings to their care. serious implications. For example, in a Increasingly, they are willing to be 2015 clinical trial, an AI app was used to monitored wirelessly for their conditions, predict which patients were likely to and to consider receiving traditionally develop complications following hospital-based medical treatments, such as pneumonia, and therefore should be chemotherapy, at home. Obtaining readings hospitalised. This app erroneously from devices like electrocardiograms, instructed doctors to send home patients pacemakers or defibrillators, which with asthma due to its inability to take generally requires an in-person visit, can contextual information into account. The performance of symptom checker apps reflect and reinforce biases in the data used using AI, has been questioned. For to train them. Concerns have been raised example, it has been found that about recommendations from apps might be the potential of AI to lead to discrimination overly cautious, potentially increasing in ways that may be hidden or which may demand for unnecessary tests and not align with legally protected treatments. characteristics, such as gender, ethnicity, disability, and age. The House of Lords • TRANSPARENCY AND Select Committee on AI has cautioned that ACCOUNTABILITY datasets used to train AI systems are often It can be difficult or impossible to poorly representative of the wider determine the underlying logic that population and, as a result, could make generates the outputs unfair decisions that reflect wider produced by AI. Some AI is proprietary and prejudices in society. The Committee also deliberately kept secret, but some are found that biases can be embedded in the simply too complex for a human to algorithms themselves, reflecting the understand. Machine learning technologies beliefs and prejudices of AI developers. can be particularly opaque because of the Several commentators have called for way they continuously tweak their own increased diversity among developers to parameters and rules as they learn. This help address this issue. The benefits of AI creates problems for validating the outputs in healthcare might not be evenly of AI systems, and identifying errors or distributed. AI might work less well where biases in the data. The new EU General data are scarce or more difficult to collect Data Protection Regulation (GDPR) states or render digitally. This could affect people that data subjects have the right not to be with rare medical conditions, or others who subject to a decision based solely on are underrepresented in clinical trials and automated processing that produces legal or research data, such as Black, Asian, and similarly significant effects. It further states minority ethnic populations. that information provided to individuals when data about them are used should LIMITS OF AI include “the existence of automated decision-making, (...) meaningful AI depends on digital data, so information about the logic involved, as inconsistencies in the availability and well as the significance and the envisaged quality of data restrict the consequences of such processing for the potential of AI. Also, significant computing data subject”. However, the scope and power is required for the analysis of large content of these restrictions - for example, and complex data sets. While many are whether and how AI can be intelligible - enthusiastic about the possible uses of AI in and how they will apply in the UK, remain the NHS, others point to the practical uncertain and contested. Related questions challenges, such as the fact that medical include who is accountable for decisions records are not consistently digitised across made by AI and how anyone harmed by the the NHS, and the lack of interoperability use of AI can seek redress. and standardisation in NHS IT systems, digital record keeping, and data labelling. • DATA BIAS, FAIRNESS, AND There are questions about the extent to EQUITY which patients and doctors are comfortable Although AI applications have the potential with digital sharing of personal health to reduce human bias and error, they can data.40 Humans have attributes that AI also systems might not be able to authentically possess, such as compassion. Clinical questions about whether and how ethical practice often involves complex judgments values or principles can ever be coded or and abilities that AI currently is unable to learnt by a machine; who, if anyone, should replicate, such as contextual knowledge and decide on these values; and whether duties the ability to read social cues. There is also that apply to humans can or should apply to debate about whether some human machines, or whether new ethical principles knowledge is tacit and cannot be taught. might be needed. Claims that AI will be able to display autonomy have been questioned on grounds CONCLUSION that this is a property essential to being human and by definition cannot be held by We reviewed the motivation of using AI in a machine. healthcare, presented the various healthcare data that AI has analysed and surveyed the CHALLENGES FOR GOVERNANCE major disease types that AI has been deployed. We then discussed in details the AI has applications in fields that are subject two major categories of AI devices: to regulation, such as data protection, ML and NLP. For ML, we focused on the research, two most popular classical techniques: and healthcare. However, AI is developing SVM and neural network, as well as the in a fast-moving and entrepreneurial modern deep learning technique. We then manner that surveyed the three major categories of AI might challenge these established applications in stroke care. frameworks. A key question is whether AI should be regulated as a distinct area, or A successful AI system must possess the whether different areas of regulation should ML component for handling structured data be reviewed with the possible impact of AI (images, EP data, genetic data) and the in mind. Further challenges include the NLP component for mining unstructured need to ensure texts. The sophisticated algorithms then that the way AI is developed and used is need to be trained through healthcare data transparent, accountable, and compatible before the system can assist physicians with with disease diagnosis and treatment public interest, and balanced with the desire suggestions. to drive UK innovation. Many have raised the need for researchers, healthcare The IBM Watson system is a pioneer in this professionals, and policy-makers to be field. The system includes both ML and equipped with the relevant skills and NLP modules, and has made promising knowledge to evaluate and make the best progress in oncology. For example, in a use of AI. cancer research, 99% of the treatment recommendations from Watson are THE FUTURE OF AI coherent with the physician decisions. Furthermore, Watson collaborated with In the future, it is likely that AI systems will Quest Diagnostics to offer the AI Genetic become more advanced and attain the Diagnostic Analysis. In addition, the ability to system started to make impact on actual carry out a wider range of tasks without clinical practices. For example, through human control or input. If this comes about, analysing genetic data, Watson successfully some have suggested that AI systems will identified the rare secondary leukaemia need to learn to ‘be ethical’ and to make caused by myelodysplastic syndromes in ethical decisions. This is the subject of Japan. much philosophical debate, raising Stroke is a chronic disease with acute system. Current healthcare environment events. Stroke management is a rather does not provide incentives for sharing data complicated process with a series of clinical on the system. Nevertheless, a healthcare decision points. Traditionally clinical revolution is under way to stimulate data research solely focused on a single or very sharing in the USA. The reform starts with limited clinical questions, while ignoring changing the health service payment the continuous nature of stroke scheme. Many payers, mostly insurance management. Taking the advantage of large companies, have shifted from rewarding the amount of data with rich information, AI is physicians by shifting the treatment volume expected to help with studying much more to the treatment outcome. Furthermore, the complicated yet much closer to real-life payers also reimburse for a medication or a clinical questions, which then leads to treatment procedure by its efficiency. better decision making in stroke Under this new environment, all the parties management. Recently, researchers have in the healthcare system, the physicians, the started endeavours along this direction and pharmaceutical companies and the patients, obtained promising initial results. have greater incentives to compile and exchange information. Similar approaches Although the AI technologies are attracting are being explored in China. substantial attentions in medical research, the real-life implementation is still facing obstacles. The first hurdle comes from the regulations. Current regulations lack of REFERENCES standards to assess the safety and efficacy of AI systems. To overcome the 1. IBM Research (5 January 2017) difficulty, the US FDA made the first IBM 5 in 5: with AI, our words will attempt to provide guidance for assessing be a window into our mental health. AI systems. The first guidance classifies AI 2. STAT (5 September 2017) IBM systems to be the ‘general wellness pitched its Watson supercomputer products’, which are loosely regulated as as a revolution in cancer care. It’s long as the devices intend for only general nowhere close. wellness and present low risk to users. The 3. British Academy and Royal Society second guidance justifies the use of real- (2017) Data management and use: world evidence to access the performance governance in the 21st century. of AI systems. Lastly, the guidance clarifies 4. See www.nuffieldbioethics.org the rules for the adaptive design in clinical 5. https://svn.bmj.com/content/2/4/230 trials, which would be widely used in assessing the operating characteristics of AI systems. Not long after the disclosure of these guidances, Arterys’ medical imaging platform became the first FDA-approved deep learning clinical platform that can help cardiologists to diagnose cardiac diseases. The second hurdle is data exchange. In order to work well, AI systems need to be trained (continuously) by data from clinical studies. However, once an AI system gets deployed after initial training with historical data, continuation of the data supply becomes a crucial issue for further development and improvement of the