AI, deep learning, and the coming revolution in health care
FAITH LAB TECH BYTES || ARNOLD SCHUCHTER
JUNE 10, 2018
Around the world there is a revolution going on in the use of AI and deep learning in healthcare. In the next few years we will see a profound and lasting transformation in medical diagnosis of all types of diseases and ailments.
A few years ago, there were only a few dozen AI startups focused on healthcare. Today, there are hundreds of healthcare-related AI startups. Big tech, including Google and Amazon, and universities across the world, notably for example, Stanford and MIT, have the best and the brightest dedicated to applying AI, machine learning and deep learning to medical R&D that addresses the most important problems in medicine.
Most doctors and other healthcare professions understandably are wondering “what’s next” and how to prepare for a tomorrow in which deep learning is both predictably and unpredictably transforming their profession.
First of all, what is: AI—simply software then enables computers to mimic (super)human intelligence; machine learning—a subset of AI that uses statistical techniques to enable computers to improve at tasks as they learn from experience; and deep learning—a subset of machine learning composed of algorithms that permit software to train itself to perform tasks by consuming and processing vast amounts of data (“Big Data”). Bottom line: all of this abstruse AI technology enables identifying patterns and solving problems by connecting mind-boggling numbers of dots.
In the case of disease, visualize the dots to be connected as symptoms and related disorders. Further visualize deep learning algorithms that, inexhaustibly and with endless practice, get better and better every nanosecond at diagnosing, learning from mistakes, and learning whether not just one but countless patients respond positively to treatments or develop new symptoms indicating that their initial diagnosis was wrong.
Looking at AI and its technology offshoots in this way makes it clear that diagnosing illnesses is a perfect mission and set of tasks. In addition, AI is poised to make a profound and lasting impact on healthcare because of three very potent trends: development of powerful new chips that run the software are getting much faster, cheaper and more energy efficient; the amazing software itself is getting exponentially more sophisticated as innovation explodes; and the digitization of healthcare is supplying vast amounts of the data and images necessary to train the algorithms of AI and its offshoots.
Unfortunately, perhaps, physicians have to sleep. AI systems do not. Physicians typically will see and study thousands of MRI images in a lifetime. Ho hum, algorithms can see trillions in virtually the blink of an eye. And even the best physicians can make errors, perhaps occasionally from fatigue. No so with AI that uses deep learning, for example, to review countless MRI scans and other images to detect lung nodules that may be malignant. That already is happening around the world in university centers of radiation oncology research, like at the University of California in San Francisco, where algorithms have been consistently outperforming radiologists.
When we use the term “revolutionary” to describe what AI, machine learning and deep learning are doing to address critical problems in disease diagnosis and treatment, this is not hype or an exaggeration. Until recently, we have had diagnostic computer programs that use a series of predefined assumptions about disease-specific features for each part of the human body. Consequently only a limited set of diseases could be identified and often these programs resulted in poor diagnostic performance and failed to reach widespread clinical adoption.
In contrast, deep learning can handle a broad spectrum of diseases in any and all parts of the entire body using all imaging modalities (X-rays, CT scans, etc.). Furthermore, hospitals across the world are adopting this technology and upgrading their computer hardware and software in order to join the incipient AI-driven “revolution” in disease diagnosis.
Thanks to AI and machine learning, the field of medical genetics (genomic medicine), that involves the diagnosis and management of hereditary disorders, has joined the “revolution” in what is the rapidly emerging medical specialty of predictive medicine.
A goal of genomic medicine is to determine how variations in the DNA of individuals can affect the risk of different diseases and to find causal explanations so that targeted therapies can be designed. With the help of a large-scale sources of Big Data, machine learning is proving that it can help researchers model the complex relationships between DNA and key molecules in a cell that may be associated with disease risks.
What this means is that AI and deep learning are helping researchers usher in a new era of effective genomic medicine and predictive medicine. For example, predicting the expected outcome of patients diagnosed with cancer is a critical step in treatment. Advances in AI-enabled genomic and imaging technologies provide physicians with vast amounts of data but prognostication remains largely subjective. This leads to much less reliable clinical treatment.
More about the subject of genomic medicine and its technology, the ways that deep learning can be used to predict the survival of patients using alternative treatment methods, will be deferred for another day. Suffice it to say here that the use of deep learning technology results in a much higher degree of prognostic accuracy than human expertise, prediction of patient outcomes and survival rates, all of which highlights the emerging role of deep learning in precision medicine and predictive medicine.
Precision and predictive medicine have the benefit of significant advances in the training of deep learning algorithms for medical imaging applications. Here again researchers in this nascent field are heralding advances in deep learning for disease prediction. In the future, using AI-powered tools radiologists will not have to spend large amounts of time analyzing radiology images for disease identification and treatment. What does it mean for radiologists? Like other medical professionals, AI only automates part of the radiologist’s work. But, as the “revolution” in medical diagnosis and treatment with the use of AI unfolds, machine learning systems will make a significant difference in determining the correct treatment options.