Robust healthcare systems require skilled labor, and there’s evidence we’re short on medical experts. The Association of American Medical Colleges predicts a shortage of up to 139,000 physicians in the U.S. by 2033. Globally, there’s a projected shortfall of up to 10 million nurses by 2030, warns the International Council of Nurses. The world needs to start training new healthcare professionals—and to invest in technology that can help limit the impact of scarce medical workers.
Much of this technology will use Artificial Intelligence (AI) to confront the complexities of human health. To be clear, AI is no replacement for the medical workforce. But it can support medical professionals in powerful ways; in the hands of doctors, nurses, and technicians, AI is improving patient care across the globe. Here’s what everyone—from healthcare providers to policy-makers to patients—should know about artificial intelligence in healthcare.
Artificial intelligence is computer technology that simulates human actions. Humans are good at two things that computers usually aren’t: thinking and learning. With AI, machines are capable of learning from large stores of data, or even from human behavior. They can use that learning to make decisions—a form of “thinking.” These AI capabilities allow computers to solve real-life problems, rather than formal, rule-based problems—and one of the most common characteristics of real-life problems is that they are hard to solve.
As this differentiation between “real-life” and “formal” problems suggests, the key difference between AI computing and a more traditional model relates to the type of logic these systems use. Two types of logic concern us here:
Fuzzy logic leads to the creation of new algorithms capable of far greater complexity than binary computing. In the healthcare field, a few key AI technologies are emerging as useful tools. Prominent artificial intelligence in healthcare includes:
Natural language processing is an AI technology that allows machines to “understand” and interpret human language. That capability leads to many language-based capabilities. For example, some NLP systems scan news articles and then summarize them, creating instant article abstracts. Before the advent of AI, only humans could perform an interpretive and creative task like this. So, how do you build an NLP system? You start with a core AI technology: machine learning.
Machine learning is just what it sounds like: A process by which computer models train on datasets to develop a form of knowledge or expertise. Machine learning has three parts: First, it trains through exposure to datasets. Then, it identifies patterns in that data. Using the knowledge obtained from these first two steps, AI with machine learning can make decisions—and help doctors make decisions. But to improve the AI tools that lead to greater health outcomes, computer scientists may employ a more powerful form of machine learning: deep learning.
Deep learning is a form of machine learning conducted on a particular architecture called a Deep Neural Network (DNN). These systems mimic the architecture of the human brain. When you learn a skill, you create an efficient electrochemical pathway between neurons—that’s what learning is. When you train a deep neural network, it creates efficient processing pathways between its artificial neurons.
To be clear, machine learning is a subset of artificial intelligence, and deep learning is a subset of machine learning. It’s just one way to achieve machine learning—but the neural connectivity model is capable of processing extremely large datasets and solving complex problems. In a high-risk industry like healthcare, you want the most accurate computational models available. That requires tremendous datasets. Compared to other forms of machine learning, deep learning achieves higher accuracy through its ability to train on larger datasets—and that makes it an ideal technology for artificial intelligence in healthcare.
Because AI can solve complex problems—and the human body is a complex system—this technology is improving medicine in a wide variety of applications. Here’s how AI helps healthcare providers in just a few examples.
Deep learning leads to digital diagnostic tools that can identify subtle health indicators humans might miss, particularly in the field of medical imaging—and particularly early in the course of the disease, when visual indicators are harder to spot. You can train an AI model with a database of medical images, from CT scans to X-rays to photos of skin conditions. With this training, the system will identify patterns that are difficult or impossible to see with the naked eye. When confronted with a new medical image, this AI tool recognizes those patterns and deviations from those patterns. That allows it to flag abnormal biological markers for the human healthcare team, including a percentage probability of its own accuracy—what’s called a confidence score.
This confidence score helps with medical workflow. Practitioners set their own benchmarks for diagnostic confidence scores. They can decide that, if the tool diagnoses a tumor with a 95% probability, for instance, it’s time to proceed to treatment. But if the confidence score is only 60%, doctors may instead choose further testing as the next step.
Treatment doesn’t end when a patient walks out of the hospital or consultation room. Many conditions require ongoing treatment, and few healthcare providers have the staff to ensure that patients take their medications at the right time or continue their physical therapy exercises. AI chatbots can step in to help.
These app- or web-based conversational assistants use NLP to understand patient queries, either through text or, in phone-based systems, spoken language. They can send reminders when it’s time for another dose or answer frequently asked questions with personalized detail. When a patient asks a question the chatbot can’t handle, it can recognize the need for human assistance and pass the issue on to a practitioner.
In many areas of the world, there simply aren’t enough doctors to go around. Lack of medical experts leads to worse health outcomes; in the United States, people who live in rural areas—where access to healthcare is more limited than in cities—face greater risk of death from heart disease, cancer, chronic lower respiratory disease, stroke, and unintentional injury than their urban counterparts.
Certainly, AI chatbots can help to fill the gap. Remote treatment, through videoconferencing, is another tech solution (albeit one that’s less dependent on AI). But the data analysis capabilities of AI can also identify areas that lack healthcare facilities. Governments and healthcare companies can use AI analysis models to determine locations where a new facility will serve the greatest number of people.
Surgical robots use AI to perform microsurgical procedures, creating tiny incisions with greater accuracy than the human hand can achieve. This application of AI is not just software; it also involves robotics. While this technology remains new, it’s already been proven in the field. In 2017, surgeons at the Dutch hospital Maastricht University Medical Center completed an advanced medical procedure with the help of an AI surgical robot. The system sutured blood vessels too small for human movements, down to just .03 millimeters in diameter.
Medical providers access records for each and every patient, and those files are often quite large, packed with key details that may contain clues for quick, accurate diagnoses. Given the volume of data in a medical file, certain details are easy to miss. One solution is Intelligent Document Processing (IDP), in which AI technologies like Computer Vision (CV), Optical Character Recognition (OCR), NLP, and machine learning categorize and extract key data from documents of all types.
At Nividous, we’ve trained machine learning models to identify medical documents—from historical data to medical tests to treatment reports—and to extract key information. These IDP systems provide doctors with the data they need to ask the right questions in their patient interviews. The same IDP systems can speed up record searches by identifying and categorizing every document in a patient’s file. And IDP is just one example of how hyperautomation on the Nividous platform uses AI to assist healthcare providers.
Artificial intelligence combined with Robotic Process Automation (RPA) and Business Process Management (BPM) leads to intelligent automation, in which a digital tool like the Nividous platform automates both individual tasks and complete, end-to-end processes, orchestrating work between RPA bots and human staff. Here are a few examples of Nividous intelligent automation in the healthcare industry: