- Big data is only as valuable as the applications used to process and present it – and those applications are starting to get eerily clever as developers get to grips with machine learning and artificial intelligence in the healthcare setting.
A number of new partnerships, projects, and collaborations featuring some of the most recognizable names in the technology world are taking on some of the biggest challenges in healthcare, including real-time clinical decision support, precision medicine, population health management, and even curing cancer.
With names like Microsoft, Facebook, Google, Amazon and IBM getting into the machine learning game, it’s no wonder that investment and excitement around artificial intelligence is reaching a fever pitch.
In a recent poll conducted by Silicon Valley Bank, 35 percent of healthcare leaders believe that artificial intelligence will have a major impact on healthcare starting as soon as 2017, and the industry is quickly putting its money where its mouth is.
A report from CB Insights shows staggeringly rapid growth in the AI solutions landscape, with investments reaching $2.38 billion in 2015.
By June of 2016, the cross-industry AI marketplace had scooped up nearly $1.5 billion in equity funding, putting developers and start-ups on track to break the previous year’s benchmark.
Machine learning, also known as cognitive computing, deep learning, and semantic computing, happens when computers are “taught” to analyze extremely large, dense, or detailed data using algorithms that can absorb previous outcome data to inform future results.
For example, a data scientist may wish to train a computer to recognize early-stage lung cancer in chest x-rays as quickly as possible.
Using imaging test results from patients who are already known to have had very early signs of lung cancer, the computer can look at the scans and identify the subtle patterns that likely indicate the beginnings of lung cancer. It can then use those positive and negative identifications to adjust the criteria for flagging future images from patients who have not yet received a diagnosis.
As the computer runs through more and more images, and collects more and more results about its accuracy rate – likely supplemented by the expertise of human clinicians – it can continue to refine its procedures and improve its diagnostic abilities.
This application would utilize a classification model, one of three main methodologies for designing a machine learning algorithm, according to a new report from the Precision Driven Health initiative, a joint effort of the University of Auckland in New Zealand, health IT developer Orion Health, and the Waitemata District Health Board.
The classification model creates a rule, or series of rules, that separates positive results from negative results based on the data fed to the algorithm during its training phase.
Meanwhile, a clustering model, in which the computer simply divides data based on similar characteristics, would be used to discover whether or not there are meaningful patterns present in a large data sample, such as a correlation between a certain medication and a certain negative reaction.
This approach could be used for developing risk stratification classifications for population health management, understanding the relationship between specific therapies and their outcomes, or matching socioeconomic and community factors with the likelihood of a patient experiencing a health event.
The last method, the regression model, is used to find a specific value, such as the time between a patient discharge and their likely return to the hospital, the paper says. Using historical data, the algorithm can find associations between two variables and predict future outcomes based on known criteria.
No matter what model is employed, computers can process all these types of data much more quickly and comprehensively than humans. And because they do not have wavering attention spans or forget any part of what they have experienced, machine learning holds a great deal of promise for data-driven diagnostics and decision support.
The imaging analytics example is on the high end of complexity for the current technical landscape, although tech giants like IBM, Dell, and Microsoft, as well as providers like Intermountain Healthcare, are throwing their weight behind its incredible potential.
Other aspects of machine learning, like natural language processing, are somewhat more familiar to healthcare providers, and are already heavily used for clinical analytics, ICD-10 coding, clinical documentation improvement, and converting PDFs of care summaries or other reports into text that can be manipulated.
Artificial intelligence may also play a significant role on the consumer side of healthcare, as it already does in the daily lives of patients who use virtual assistants like Siri, Cortana, or Alexa on their smartphones and other devices.
As healthcare becomes more consumer-focused due to rising out-of-pocket costs and an emphasis on patient-centered care, providers are searching for new ways to engage and manage patients without increasing the clinical workload.
Artificially intelligent personalities may be able to assume some human duties like answering simple administrative questions or resolving billing issues for patients, and could be given the task of helping patients manage chronic diseases, collecting patient information and health histories, or triaging lower-level concerns to prevent workflow overloads for clinicians.
Before they can be used for any of these tasks, however, AI programs have to become significantly more sophisticated. One or two elite machine learning tools have arguably already passed parts of the Turing test, meaning they can conduct certain tasks that can fool humans into believing that other humans completed the challenge.
The majority of existing AI applications are somewhat less refined, but IBM and MIT are hoping to change that with a collaboration that will develop cognitive learning systems that can emulate the human brain’s ability to integrate audio and visual inputs into their pattern recognition capabilities.
The new IBM-MIT Laboratory for Brain-inspired Multimedia Machine Comprehension hopes to create machine learning applications that can complete tasks such as summarizing the content of a video clip, such as a telehealth visit or a webcam feed from a remote monitoring device in an elderly patient’s home.
Someday, this type of artificial intelligence could diagnose a skin rash through a video consult so a patient can avoid an office visit or flag a change in behavior that may indicate a new neurological issue during a remote check-in for a Parkinson’s patient.
"In a world where humans and machines are working together in increasingly collaborative relationships, breakthroughs in the field of machine vision will potentially help us live healthier more productive lives," said Guru Banavar, Chief Scientist, Cognitive Computing and VP at IBM Research.
"By bringing together brain researchers and computer scientists to solve this complex technical challenge, we will advance the state-of-the-art in AI with our collaborators at MIT."
IBM is certainly one of the leaders in the field of cognitive computing, and has made a special effort to make progress in the healthcare industry.
The formation of IBM Watson Health in 2015 was a clear signal that patient care is a top priority for the company, and a variety of acquisitions, partnerships, and projects in the field are using machine learning to address issues in the realms of precision medicine, clinical decision support, workplace wellness, and cancer diagnostics.
Cancer is also a focus for Microsoft, which has been delving into the world of personal AI assistants with its Cortana software. At the venerable technology behemoth’s global research labs, programmers, computational biologists, machine learning specialists, and clinicians are collaborating to “solve cancer,” says a recent announcement penned by Allison Linn.
The company is exploring two promising pathways for cancer breakthroughs, rooted in the explosion of new discoveries related to genomics and precision medicine.
“One approach is rooted in the idea that cancer and other biological processes are information processing systems,” Linn wrote. “Using that approach the tools that are used to model and reason about computational processes – such as programming languages, compilers and model checkers – are used to model and reason about biological processes.”
“The other approach is more data-driven,” she continued. “It’s based on the idea that researchers can apply techniques such as machine learning to the plethora of biological data that has suddenly become available, and use those sophisticated analysis tools to better understand and treat cancer.”
One of the goals is to make it easier for clinicians to diagnose cancer and personalize treatments based on knowledge distilled from millions of lines of text in medical journals and other documents. Another is to understand the biological processes that flip the cellular switch from normal to malignant.
Machine learning also could create a virtual treatment environment, where simulations based on an individual’s genome, clinical data, and other information can predict which available therapies have a higher or low likelihood of success before a needle breaks skin or a pill passes the patient’s lips.
Developing these futuristic capabilities will likely require many years of research and expertise from across multiple industries, however.
The retail and consumer technology industries are far ahead of the healthcare sector when it comes to leveraging artificial intelligence and machine learning capabilities, but it may bode well for providers and patients that most of these companies are looking to apply their foundational big data knowledge to as many disciplines as possible.
Microsoft and IBM, along with Google, Facebook, and Amazon, are hoping to accelerate this collaborative approach to knowledge with their new Partnership on Artificial Intelligence to Benefit People and Society, an organization geared towards developing and disseminating best practices for advanced machine learning techniques.
“We’re in a golden age of Machine Learning and AI,” said Ralf Herbrich, Director of Machine Learning Science and Core Machine Learning at Amazon. “As a scientific community, we are still a long way from being able to do things the way humans do things, but we’re solving unbelievably complex problems every day and making incredibly rapid progress.”
Google is entering the partnership via its DeepMind subsidiary, which made waves in the healthcare industry earlier this year thanks to the questionable privacy issues surrounding a patient data sharing agreement in the United Kingdom.
“Google and DeepMind strongly support an open, collaborative process for developing AI. This group is a huge step forward, breaking down barriers for AI teams to share best practices, research ways to maximize societal benefits, and tackle ethical concerns, and make it easier for those in other fields to engage with everyone’s work,” said Mustafa Suleyman, Co-Founder and Head of Applied AI at DeepMind and Greg Corrado, Senior Research Scientist at Google in a statement.
Ethical concerns, including privacy issues, are likely to be a major factor during the development of AI for healthcare, due to the highly sensitive nature of patient data and the wide array of applications for the insights that may come out of machine learning algorithms.
There are numerous questions to be answered before algorithmic assistants can dispense the ideal cancer treatment before a patient even starts showing symptoms, but the potential for machine learning and AI to revolutionize nearly every aspect of healthcare seems clear.
As providers and developers continue to collaborate and innovate across the booming machine learning industry, healthcare providers and patients are likely to see a long line of promising breakthroughs fueled by more big data than ever before.