Coming to a mobile device near you: machine learning and healthcare.
Machine learning powers more and more actions in our daily lives. Search engine results, email spam filters, product recommendations on shopping sites like Amazon, credit card company’s screening for fraudulent purchases, and Siri’s voice recognition software all function thanks to machine learning. Every day, machine learning expands into other areas and applications.
In 1959, the computer gaming and artificial intelligence pioneer Arthur Samuel defined machine learning as a “Field of study that gives computers the ability to learn without being explicitly programmed.” Instead of relying on programming, a computer applies algorithms to a data set, learns from the results, and repeats. In short, the computer completes an indefinite series of tiny experiments that “teaches” the computer and reveals hidden insights and patterns within the data. The only limitations are the quantity and quality of the data set, and the appropriateness of the algorithms.
This machine learning process creates models for application on additional data or larger data sets, and does so much faster than humans are capable. For instance, in Indianapolis, a joint Regenstrief Institute and Indiana University research project demonstrated that machine learning could identify cancer cases from text-based pathology reports as effectively as a human reviewer, only much faster. Dr. Shaun Grannis, senior study author and interim director of the Regenstrief Center, said in an Indiana University press release, “We have come to the point in time that technology can handle this. A human’s time is better spent helping other humans by providing them with better clinical care.”
Prior to the pathology results study, Dr. Grannis and the Regenstrief Institute helped create Indiana’s Public Health Emergency Surveillance System which now detects communicable disease outbreaks roughly a week earlier than previously possible via human reporting. If formally implemented in pathology screening, machine learning could eliminate the existing months-long cancer pathology backlog and speed up transfer of cancer statistics from healthcare providers to public health departments. The results? More timely access to decision-making information and allocation of resources.
Another impressive use of machine learning involves examining possible correlations between biomarkers and the incidence of depression. An Australian study led by the IMPACT Strategic Research Center analyzed data from the 2009–2010 U.S. National Health and Nutrition Examination Study in an attempt to link depression with one or more biomarkers such as blood cell counts, cholesterol levels and nutrient levels. Using machine learning, 21 of the 67 studied biomarkers were initially linked to the occurrence of depression, but further algorithm refinement determined only three were true signals — red cell distribution width, serum glucose, and total bilirubin.
Now, imagine the findings of these two studies, pathology and biomarker analysis, were developed into a mobile technology. Perhaps a pathology lab uses an app on a mobile tablet to record its results, which automatically applies the machine learning model, identifying a cancer diagnosis then triggering the information’s transmittal to the public health department. Or, imagine a physician uses a patient’s blood test results, a mobile tablet, and a biomarker app powered by a machine learning model to identify a patient at risk for depression during his or her annual health screening. In an instant, the physician has results to provide preventative healthcare, advising the patient how to improve their physical, and indirectly, their mental health.
The power of machine learning need not be limited to the internet, big data, big research, and powerful computers. Thanks to more powerful computing and off-network machine learning capacities, the mobile machine learning possibilities are powerful.
— Embrace hope.
What are your ideas on how machine learning on a mobile device could improve healthcare?
— — — — — — — — — — — — — — — — — —