The Definitive Guide To AI In Healthcare
Artificial intelligence (AI) in healthcare is finding numerous applications across the different spectrums of healthcare and is all set to transform the way we diagnose and treat illness in the days to come.
The juxtaposing of big data with AI via machine learning and natural language processing has led to development of tools to improve clinical care, advance medical research and improve overall care efficiency.
This comprehensive guide is all you need to know about Artificial Intelligence in healthcare, its working, applications, future healthcare technology trends and challenges in its adoption in the healthcare industry.
How is the healthcare segment transforming?
In earlier times, healthcare was primarily focused on delivering evidence based care. It majorly involved diagnosis and cure of diseases and the entire system was solely involved in the delivery of healthcare products and differentiation was solely through product innovation.
The current decade is seeing the shift from this traditional outlook and is rapidly moving towards an outcome based care based on real time data collected from different healthcare platforms and wearable devices.
The collaboration of big data, AI and robotics in the health sector will result in development of intelligent medical solutions. These would be used to provide both evidence and outcome based care with the focus on preventative care.
What is AI?
Artificial intelligence is the use of specific algorithms to train computers for completing specific tasks by processing a large amount of data and by recognizing specific patterns in the data.
These algorithms make it possible for the machined to learn from experience, compute the input data and process it and perform tasks in an almost human like manner. It thus enables the machines to mimic the cognitive functions typical to the human mind like problem solving, reasoning and learning.
Reasons for proliferation of AI
There has been a surge in proliferation of AI in healthcare recently which can be attributed to a number of factors.
Explosion in the amount of healthcare data: From individual health stats collected from a number of medical devices and fitness wearable, to electronic patient health records stored in the hospital database, the amount of healthcare data being collected and stored has become enormous. The computation of this vast amount of data isn’t humanly possible and stems the need for artificial intelligence to process this data. The clinical data available including images and lab reports are instrumental in training the AI systems before they can be deployed.
Transformation in the data computing capacities: The computers are now equipped to process data at a phenomenal rate. The evolution of internet networks now allows data transfer at lightning fast speeds. With increased adoption of cloud computing in healthcare, patient data is no longer saved on the internal servers but is rather hosted on the cloud. This allows for greater collaboration and improved data accessibility. There is also a significant advancement in the analytics techniques which act as a propeller for AI in healthcare as well.
Change in focus of healthcare delivery: The traditional hospital setups were meant for management of acute conditions which were typically high in severity and would require days or weeks of care. The increased prevalence of chronic conditions and increase in elderly population globally demands of facilities and equipment catering towards long term care. AI is a step in the right direction to cope with this challenge while keeping the healthcare costs to minimum.
Shift in health ownership: With an increasing number of people relying on wearable medical devices and other fitness variables, the individuals are adopting a more comprehensive role in looking out for their own health. Healthcare mobility has resulted in people carrying their medical records in their own smartphones for easy accessibility. This shift in perspective requires the healthcare system to also transform to the individual needs and adopt a more holistic, data driven approach.
Applications of AI
AI has already entered our day to day lives in such a manner that we don’t even notice it. From chatbots on eCommerce websites to voice assistants on our smartphones, AI consultants are rapidly making machines learn what humans do and enabling them to do those tasks more effectively and rapidly.
In the healthcare sector too, AI tools have found numerous applications across the different stages
1. Preventive Medicine
“Health is not merely the absence of disease or infirmity, it is a state of complete physical, mental and social well being” – World Health Organization
AI with the use of connected medical devices and IoMT has the biggest potential role in ensuring that people stay healthy.
For the general population: AI is already helping people in taking care of their own health by maintaining a healthier lifestyle. An increasingly large number of people are choosing to adopt fitness and other health wearables to track their health statistics on a daily basis.
Collection and analysis of this health data and its supplementation with health information provided by the patient through health apps has the potential to offer a unique perspective into the individual and population health.
For medical practitioners: AI technology is providing the healthcare professionals an insight into the day to day patterns and needs of the people they care for. This enables them to provide better guidance, feedback and support to their patients.
The data generated and collected is not computable by humans in isolation. Use of AI algorithms by custom software development companies to make sense of the vast data not only saves precarious human time and efforts but also makes the process more efficient. Software development companies can create dashboards that can present the data visually that can save a ton of time from manual data-crunching.
For researchers: Collection of crowd-sourced medical data is something that the companies focusing on healthcare research are focused on. Data from various mobile devices is being pooled and aggregated in order to gain access to live health stats.
Real life applications
- Under armour: The US based apparel manufacturer works on a cognitive coaching system that provides customized advice regarding fitness, sleep and nutrition. It works on using data from the fitness community, research institutions which it analyses uses IBM’s Watson.
- Lumo lift: This posture monitoring device works by sending out alerts to the user whenever it detects incorrect posture while standing or sitting.
- Apple’s researchkit: Apple is focusing on treatment of Parkinson’s disease and Asperger syndrome by letting the users access interactive apps for pooling data collected by its facial recognition software. It then plans to document the change in this data over time.
2. Early Detection of Diseases
The proliferation of consumer wearables combined with the computation power of AI has enabled doctors and other caregivers to better monitor the patient and detect the potentially life-threatening episodes at an earlier, more treatable stage. Here are some applications of the same that are currently being used in the healthcare space.
Detection of Cardiac conditions: Fitness and other health wearable devices can be used to detect not just the heart rate but also monitor patient’s ECG. This is instrumental in detection and earlier diagnosis of underlying cardiac conditions.
Detection of breast cancer: AI is being used currently to analyse mammographs. It has been discovered that the analysis rate is 30 times faster than that of a human and has an accuracy of 99 percent. This not only reduces the chances of possible misdiagnosis, it also reduces the need to perform invasive biopsies to reach the diagnosis.
Infection trend prediction: Sepsis is one of the leading causes of hospital deaths in the US and the diagnosis doesn’t usually happen until the development of organ failure. Application of AI in detection of sepsis can go a long way in decreasing the patient mortality rate. Work is currently underway to develop a computer algorithm that analyzes the patient vitals and metabolic levels from the patient’s EHRs to detect if they have a likelihood of getting sepsis.
Disease prevalence trends: The patients are increasingly relying on search engines like Google to check their symptoms online before paying a visit to the doctor. The use of AI to monitor this search and draw conclusions from it can lead to early intervention in possible disease outbreak in the population. Google had tried to do this with Google flu trends back in 2008 but failed due to lack of streamlined data and numerous inconsistencies. With progress in computing and AI in the last decade, this can turn to be a big asset in early detection of infectious diseases and preventing their outbreaks.
Turning Electronic health records into risk predictors: Patient’s medical records are a data goldmine but sorting through the data and coming up with useful results is a task that would result in lot of human time and efforts going waste. This is where AI’s computing power comes to play. EHR analytics tools that employ deep learning techniques have been used to unearth valuable patient data by using risk scoring and stratification tools resulting in predictive analytics that lets the machine learning algorithms to determine the patient’s risk to acquire chronic diseases.
Real life applications
- Apple’s generation 4 iWatch has the ability to monitor the wearer’s ECG similar to a single lead electrocardiogram and detect any abnormalities. It also sends out distress messages to the family members and the emergency services when its gyroscope detects a potential fall.
- Houston Methodist Research Institute in Texas is working on analysis of mammographs which would result in early and non-invasive diagnosis of breast cancer in women with a reduction in misdiagnosis.
3. Efficient Diagnosis
Artificial Intelligence uses both structured and unstructured data for obtaining its results. The structured data includes genomic studies, images (radio diagnostic and pathological), readings and recordings form medical devices etc. This data is then clustered using machine learning techniques to infer diagnosis and possible disease outcomes.
The unstructured data can be in the form of the physician’s notes, patient medical records in the form of EHRs, lab reports, discharge summaries etc. AI makes use of natural language processing for extracting the relevant information from the sets of unstructured data in order to assist in clinical decision making, alerting treatment arrangement, monitoring adverse reactions etc.
Use of AI can help in coming up with the diagnosis in a more effective manner by using both structured and unstructured data at a much faster rate. The major advantage of AI for reaching a diagnosis is that all its decisions are solely evidence based and free of cognitive bias which may result in human diagnosis. Let’s take a look at the applications of AI in diagnosis of diseases.
Diagnosis using radiographs: Use of AI for image analysis of radiological images obtained from MRI machines, CT scans and X-rays not only resulted in diagnosis that were at par with a human radiologist, the results obtained were much faster as well. Application of AI in radiodiagnosis meant to be an adjuvant to the radiologist, who can use AI for routine cases and utilize his/her resources for more complicated cases.
Use of AI in oncology: AI is being trained to recognize and identify skin lesions for diagnosis of skin cancer based on facial recognition techniques. As mentioned above, it is used in diagnosis of breast cancers by mammograph analysis. Virtual biopsies by using AI are harnessing the image based algorithms to make advances in the field of radionomics. This would allow clinicians to develop a more accurate understanding of tumour behaviour as a whole and give them the ability to better define the aggressiveness of tumours and select the treatment that would result in the best outcomes
Application of AI in pathology: Pathological diagnosis involves examination of the section of tissue under a microscope. Incorporation of deep learning to train an algorithm for image recognition would provide more accurate diagnosis when combined with human expertise. Analytics of the large digital images at the pixel level can help in detection of pathological lesions which may escape the human eye and lead to a more efficient diagnosis.
Real life applications
- Microsoft’s InnerEye initiative: it uses machine learning for analysis of radiographic images and establishing image diagnostic tools. The results can be utilized for planning of radiotherapy, and precise surgical planning and navigation.
- Stanford University: The researchers at Stanford university are working on deep learning using convolutional neural networks to train an algorithm to diagnose skin cancer. The best thing is that it works on smartphones and can be used to detect melanomas from over 2000 different skin lesions.
- Harvard Medical school: Researchers from Harvard medical school and Beth Israel Deaconess Medical Centre have developed AI powered systems to make pathological diagnosis more accurate by using deep learning and machine learning. The algorithm makes use of speech and image recognition to interpret pathological images and trains computers to differentiate between cancerous and non cancerous lesions. Combination of this algorithm with the pathologist’s work led to a 99.5 percent accuracy rate.
4. Medical Decision Making
The use of AI algorithms to support clinical decision making, early alerting and risk scoring ensures delivery of quality clinical care. AI systems do not suffer from human deficits like decision fatigue and alarm fatigue so use of these to determine the clinical workflow and its management would result in more efficient patient care.
Administrative workflow management: Leveraging AI for automation of administrative workflow via custom software development ensures that the care providers like doctors and nurse practitioners save time on routine tasks and can prioritize on urgent matters instead. Management of routine tasks like entry of medical notes in patient’s charts can be done using voice to text transcriptions that can save valuable time.
Predictive analysis: The patient data collected in form of electronic medical records and those obtained from the wearable devices gives the physician access to valuable data about the patient as well as the population cohort the patient belongs to. Computation of this data using AI algorithm helps develop the patient profile and build predictive models to effectively anticipate, diagnose and treat the diseases.
Clinical decision making: Having access to complete patient data is a boon for clinical decision making. It also results in development of a treatment model that caters to the needs of the individual patient which is different from the generic approach traditionally adopted.
Real life applications
- Google’s DeepMind: This technology uses machine learning to build learning algorithms into neural networks that mimic human brain. It aims to assist clinicians in medical decision making and help them go from tests to treatments at a faster rate.
- IBM Watson: Memorial Sloan Kettering (MSK)’s oncology department has partnered with IBM Watson for providing AI based solutions for cancer treatment and research. IT has decades worth of cancer data which can be utilized for providing treatment ideas to doctors dealing with future cases.
- PwC’s Bodylogical: Bodylogical applies prescriptive analytics tools to fuse science with analytics. It results in creation of digital twins (a virtual version of the user that models human body) to predict disease progression, monitor health status, determine appropriate lifestyle changes and thus help in lowering overall healthcare costs.
- Nuance: The company works on computer assisted physician documentation (CAPD) to provide clinical documentation improvement (CDI). These AI powered solutions cut down the documentation time and improve the reporting quality.
AI can help clinicians in having a more comprehensive approach for disease management, result in better coordination of care plans and ultimately help patients to become more compliant with their long-term treatment programs. Here are some of the applications of the same.
Virtual nursing assistants: Work is currently under progress to use AI for development of virtual nurses that would be available at the patient’s bedside throughout the treatment. They are used to monitor the patient stats and provide answers to the routine questions. They establish channels of communication between the doctors and the patients at regular intervals and thus help prevent unnecessary hospital visits, saving costs. The nurse avatars are voice based and use verbal communication to converse with the patients.
Voice to text transcriptions: A significant amount of time is spent by healthcare providers in entering medical or surgical notes in patient’s health records. AI-enabled voice to text transcription of these notes would increase the time spent in patient care and improve clinical effectiveness.
Precision Medicine: Having the relevant patient data at the clinician’s disposal is a step in the right direction for development of precision medicine. It enables the physicians to take medical decisions catering to the individual patient and create specific treatment plans for each patient.
Real life applications
- Sensely is a virtual nursing assistant in the form of an avatar capable of remote patient monitoring. It uses speech recognition software on its proprietary classification engine to listen to the user’s queries and deliver an appropriate response.
- Hello Rache is a live virtual assistant that transcribes, assists and performs the routine administrative tasks for medical professionals.
- Royal Philips enables remote patient monitoring of by use of sensors and data analytics. The data is then streamed to remote care team to decide upon the treatment plan.
- Pharmacogenomics: The drug interactions and efficacy vary from person to person and are influenced by genetic variations. Pharmacogenomics aims to understand the effect of these variations on individual responses to medications. Use of AI to sift through the large amount of data produced by genomic studies can help give useful insights into drug delivery mechanisms.
- Drug discovery and drug combination analysis: Drug research takes numerous years and millions of dollars before the clinical trials and marketing of the drug can take place. Use of AI to streamline the drug discovery and drug repurposing processes has the potential to significantly boost new drug development, reduce the time to market for the drugs and also reduce their costs.
Real life applications
- In a clinical trial in 2016, researchers made use of AI to come up with a mathematical formula for determining the dose of immunosuppressants to be administered to patients slated for receiving organs.
- Pharma.AI is a bioinformatics company that uses AI to research drug discovery programmes for cancer, Alzheimer’s, Parkinson’s, and other ageing and age‑related health issues.
- Atomwise in a recent research used AI to scan existing medicine that could be redesigned to fight against Ebola virus. In a span of just one day it found two medicines with the potential to decrease infectivity, an analysis that would typically take a huge amount of time.
7. Robotics and Chatbots
- Robot assisted surgery: Cognitive surgical robots use AI to use information for previous surgeries to improve upon the surgical techniques. Data from pre op records are integrated with the operating metrics to improve upon surgical outcomes. These surgeries are minimally invasive and the robot assisted instrument precision results in 21 percent reduction in patient’s length of stay post op.
- Autonomous Robotic surgeries: While currently a thing restricted to science fiction, robot only surgeries can be a real thing in the future. Use of machine learning to combine motor pattern recognition and visual data interpretation can result in extension of surgeon’s dexterity to robots and make autonomous robotic surgery a reality. The robot only surgery is currently limited to surgeon’s controlling the robots remotely via a computer but that can transform in the days to come.
- Auxiliary robots: These are robots that find clinical application in a variety of arenas including patient care, nursing care, laboratory use, and care of elderly and debilitated patients.
- Chatbots: Chatbots are AI-powered algorithms, built by healthcare app developers, capable of carrying out the basic conversations with the end users. They have the potential to become the first point of contact for primary care. The severity of the query is determined and the chatbots ether resolve the issue or escalate it up to the physician. The widespread use of Chatbots greatly reduces the physician’s burden as well as negates the need for unnecessary drive to the healthcare providers.
Real life applications
- Da Vinci robotic systems include a wide range of clinical robots which are capable of aiding a human surgeon and can also perform operations remotely controlled by the surgeons via a computer.
- RoBear is a nursing-care robot with the ability to lift and move patients in and out of bed into a wheelchair, help those who need assistance to stand, and even turn patients in bed to prevent bedsores.
- Xenex robots are used for disinfecting the hospital with UV light to prevent the occurrence of hospital acquired infections.
- Chatbots like your.md, Izzy, Healthtap, ada health, etc are commonly being used as chatbots to avail medical information. These are downloaded as mobile applications on the user’s smartphone and use AI algorithms and natural language processing to give textual/vocal health related feedback.
Challenges to adoption of AI
The adoption of AI in the field of healthcare opens up a number of possibilities but they come with their own set of challenges as well.
1. Initial Adoption Issues
In order to attract stakeholders for investing in AI, successful case studies need to be documented and presented but in order to come up with case studies, healthcare companies need to be on board. As with any new technology, there is an initial hesitation to adoption in the market with both healthcare organizations and users having concerns to its applicability and feasibility.
2. Black Box Difficulty
Machine learning and deep learning lack the ability to give the answer to “why” questions. The logic behind the conclusion obtained isn’t justified which results in lack of confidence in the results achieved. How the system came up with a diagnosis or recommendation is an important part of the treatment plan and hence at the end of the day, the final word would be that of the physician.
3. Data Privacy Concerns
Patient health stats constitute extremely sensitive data and proper mechanisms need to be in place to ensure the safety of it from external attacks.
4. Stakeholder Complexities
Everyone in the healthcare industry, including the patients, healthcare workers, pharma companies, insurance companies, healthcare organization act as stakeholders in the adoption of AI. Resistance of the technology at any level would lead to issues with the incorporation of the technology as a whole.
5. Compliance to regulations
Patient data collection is subject to a number of laws such as HIPAA and incorporation of AI is subject to approvals from organizations such as the FDA to ensure the upkeep of the federal standards. Sharing of data across various databases in order to be analysed by AI algorithms poses challenge in terms of HIPAA compliance.