Download the authoritative guide: Cloud Computing 2019: Using the Cloud for Competitive Advantage
Healthcare sectors are increasingly adopting artificial intelligence to improve patient care and improve process efficiencies. Clearly, the use of AI in medicine has been expanding in the last few years. This is partly due to a desire by medical providers to expand their care offerings, and partly due to the maturing of artificial intelligence itself – AI has grown by leaps and bounds in the last couple years.
At this point, AI in healthcare spans many of the core areas in medicine. AI shapes – to a lesser or greater degree – everything from diagnostics to health and wellness to smart devices. In many ways, AI technology has become a "second layer" of healthcare provider. This is because AI software can adapt without human intervention, so it can "learn" to target human health needs on its own.
Not surprisingly, many top top AI companies are cashing in on the trend. With all of the investment and growth in AI technology, expect many more AI use cases for healthcare in the years ahead. Furthermore, companies can now use AI as a Service, or build their own intelligent apps using cloud-based AI services. Along with Big Data in healthcare, AI in healthcare is fast becoming a defining factor.
Let's look at what's currently happening with AI in healthcare.
Compelling Use Cases for AI in Healthcare
Antibiotics, of course, help keep people healthy. However, their pervasive use is resulting in antibiotics-resistant bacteria that kills 70,000 people per year globally. Researchers use machine learning (an AI technique) to identify genes that cause antibiotic resistance in bacteria. AI is also being used to identify pre-symptomatic patterns in electronic healthcare records (EHRs) so more and earlier alerts can be sent to healthcare providers.
Brain-computer interfaces are not a mainstream technology yet. However, there is a lot of interest in this area because brain-computer interfaces can replace other types of computer interfaces, which is particularly helpful for people with permanent or temporary disabilities. For example, AI-enabled brain-computer interfaces can help stroke patients communicate with healthcare providers soon after a stroke rather than after rehabilitative therapy.
AI has been used in cardiology for more than 20 years but its progress is slow, given the life and death consequences of heart conditions. An example of AI use is an implantable defibrillator that monitors the heart rhythms of patients at risk of a sudden heart attack. The device also administers a shock if necessary.
Over the longer term, data from wearables and implantables will be combined with Electronic Healthcare Records (EHRs) for continuous patient monitoring so doctors have more current information about their patients.
Developing nations have a different sent of problems than first world nations. First world nations are interested in ever more sophisticated forms of AI while developing nations are more concerned about providing basic services, including healthcare, to poor citizens and citizens living in remote areas. Quite often, poverty and life in remote location go hand in hand.
As a result, developing nations are using AI to provide healthcare access to those who would otherwise have no access to healthcare. Specifically, medical information pushed via a tablet to a member of the community who can read it and take appropriate action. The community representative can also use the tablet to take pictures of patients' symptoms which the image recognition system compares with similar images to diagnose the condition.
Electonic Health Records
EHRs haven't completely replaced paper yet, and even though their use is pervasive, receptionists, medical assistants and doctors must do a lot of manual entry. Here, voice recognition capabilities replace keyboards. So, instead of typing information into the system, the user can simply speak the information they want recorded in the EHR.
Video-based image recognition capabilities will likely supplement EHRs in the future because it provides additional insight into patients' conditions that AI is capable of analyzing, but humans may miss. For example, image analysis systems can tell when a patient is lying about pain, which may indicate opiate-seeking behavior.
Health and Wellness
More consumers are wearing health and fitness bands or smart watches, though there are also medical grade devices that track even more information. Such devices, depending on their design and level of sophistication can provide insight into a person's heart rate, oxygen level, sugar level, sleep patterns, breathing, gait, and more, providing healthcare providers with information they wouldn't otherwise get between appointments.
For example, a stroke patient's recovery may show improvement based on the patient's gait while the early signs of a heart attack could mean the difference between surgery and no surgery. AI recognizes patterns in the data to determine the current health status of a patient.
Immunotherapy for Cancer Treatment
Immunotherapy for cancer is not an exact science. While many immunotherapy options are available, a patient's DNA determines whether the treatment will be effective. Since AI can analyze far more information far faster than humans, it's capable of recognizing patterns in genetics strings and correlating those against immunotherapy options. The capability could result in a truly personalized approach to cancer treatment.
AI systems can analyze far more data far faster than humans, which may make them more adept at identifying medical diagnoses than doctors. For example, when a patient with a serious condition receives a diagnosis, friends and family encourage that person "to get a second opinion" since human doctors often interpret medical information differently.
AI uses historical data from hundreds, thousands, or even millions of diagnoses and then compares that with a patient's condition to diagnose the malady, to predict the progression of the malady and to recommend treatment.
Neurological healthcare deals with nervous systems disorders such as Parkinson's disease, Alzheimer's disease, epilepsy, stroke, and multiple sclerosis. AI can monitor patients with neurological disorders around the clock to see whether the patient's status is improving or declining. AI can also predict strokes and monitor seizure frequency.
Most diagnoses depend on a pathology result, so a pathology report's accuracy can make the difference between diagnosis and misdiagnosis. AI can "see" pathology results at the pixel level which can indicate the progression of cancer, for example. AI also helps humans focus on the most relevant area of a pathology image.
Various forms of radiology, such as CT scans, MRIs and X-rays provide healthcare providers with an inside view of a patient's body. However, different radiology experts and doctors tend to interpret such images differently. AI helps enable more consistent interpretations. It also helps radiologists better identify the status of a tumor or the aggressiveness of a cancer.
Hospitals are big purchasers of smart devices. The devices, which take the form of tablets and hospital equipment, exist in intensive care units (ICUs), emergency rooms, surgery and regular hospital rooms. AI supplements the medical staff by monitoring the condition of patients and alerting the proper personnel to an important status change which may have to do with oxygen levels, breathing patterns, heartbeat, blood pressure or an infection such as sepsis.
AI is used in operating rooms in an assistive capacity to narrow the considerable variance between the experience and knowledge of various doctors. An AI-enabled system is able to comb through vast amounts of data quickly to surface the information the doctor requires.
Risks of AI in Healthcare
A common issue with AI adoption in healthcare (or any industry for that matter) is that the designers and users of AI-enabled systems tend to focus more on the potential benefits than the potential risks. While it seems like everyone is talking about AI these days, few people understand the topic well. The result is that people are building and procuring systems and software that they don't completely understand.
It is the 800-pound gorilla in the room that goes unnoticed by people who have little knowledge of AI. Bias is an important topic because it makes AI systems less accurate than they could be and it may cause unintended outcomes.
Biased AI results stems from the intentional or unintentional bias of the algorithm's author or the people who collect, select, and use data. The data itself may be biased. Given the vast amount of data (big data) the healthcare industry uses and the necessity for accurate data analysis, it is important to be aware of and compensate for bias.
Wrong Decisions or Recommendations
The healthcare industry increasingly relies on AI for decision-making. The problem with a hard-coded system is it may not account for all scenarios. Self-learning systems are more flexible; however, not all systems are able to explain their results or recommendations, nor the factors that contributed to the results or recommendations. Bias may also be present in the system. As a result, it is possible an AI-powered system could make erroneous recommendations or decisions for which the healthcare provider is liable.
Doctors take and are bound to honor the Hippocratic Oath, "first, do no harm." But how does AI work with this? Most people consider AI amoral because "it's just a tool." It's the operator's use of AI that results in a moral or immoral outcome.
However, since self-learning AI systems can perceive things humans can't and they are not necessarily able to explain their reasoning or conclusions, unexpected results may occur, some of which may be unethical. Moreover, AI lacks compassion and empathy at the present time, so its decision-making processes differ from that of humans.
The Health Insurance Portability and Accountability Act (HIPAA) has strict rules about the use of healthcare information and what healthcare providers do with it. AI can malfunction as the result of machine learning training on "bad" data, algorithmic bias, or failure to maintain the system. Hackers can also compromise a system if it's not secured properly.