While wearable health technology evolves quickly in pursuit of clinical grade sensors and analysis, research into applications of artificial intelligence (AI) proceeds in universities and research centers around the globe. One of the most promising and exciting AI applications is the use of machine learning, specifically deep learning, in making sense of massive amounts of medical data sometimes referred to as “Big Data.” For example, IBM’s Watson is already helping out physicians by reading digital images from X-rays and other scans.
Scientists working at Google have made significant progress investigating how deep learning can be applied in digital pathology diagnostics of breast and prostate cancer. When biopsies are performed, pathologists are responsible for reviewing all biological tissues on a slide. Each patient typically has many slides that must be examined. This analysis poses a prodigious challenge when you consider that each slide has contains more than 10 gigapixels, digitized at 40X magnification. In the limited time available, pathologists need to scan and accurately assess each image. When different pathologists examine such images from the same patient, they disagree on the diagnosis more than half the time.
Google’s researchers used the same standard images used in training pathologists to train the deep learning algorithms. After a promising start, more training resulted in the deep learning system that produced diagnostic results that matched or exceeded the performance of pathologists. No one is ready to turn cancer diagnosis over to machines, but using AI as a prescreen or a parallel diagnostic technology holds great promise. The full study report is available here.
Work goes on building and refining methods and algorithms for machine learning approaches to big data sets of quantities beyond the resources of normal monitoring and scanning. Data from a wide spectrum of biomarker sensors, including consumer health wearables, inevitably will be checked as a master of fact, spitting out valid and reliable reports to medical personnel who can then verify and treat the diseases and conditions. It’s easy to come up with a list of potential applications, to the point that it becomes hard to think of a medical or health issue that couldn’t be better understood, treated, and perhaps cured by real-time monitoring with superlative accuracy.