Will Big Data Analytics Kickoff a New Golden Age for Radiology?
No one really knows how epiphanies come to be. Those moments of enlightenment, slivers in time, when the curtains are thrown back and there, revealed, is the hardly expected.
Maybe the facts that form those revelations rattle around our heads like the stainless steel balls in a pinball machine — flippers flapping, balls popping up, hitting the glass until — suddenly — all is quiet, the lights on the machine dim and the score thunks steadily and rhythmically upward.
That is the analog version. A digital one might take shape in the wake of enterprise imaging. Making it happen will be Big Data analytics, software that mines the data gathered in the daily course of adding information to electronic medical records.
Patterns of healthcare practice form from data that define care given to patients; lead to rules that adjust patient care protocols and are fine-tuned for effect by revisiting data conglomerated from multiple mining operations to establish a broader context. This could breed consistency where consistency is needed most, promoting wellness and more efficient delivery of care. And this would still be just the first stage of enlightenment.
Patient outcomes analyses will follow, then the implementation of new protocols. Some inevitably will involve shifts in the usage of imaging modalities, leading to clinical decision-support to guide referring physicians in the ordering of appropriate exams; leading to scan protocols best suited to answering specific clinical questions.
Radiology could do well. Armed with patient outcomes data, reimbursement could increase for procedures that increase efficiency, reduce cost and improve patient care.
Some of these early successes will likely involve the data mining of radiology reports made available through the ongoing transition from picture archiving and communication systems (PACS) to enterprise imaging. The next step in this data-driven revolution will be to link the information obtained from radiology reports to the vast stores of information contained in the pixels of medical images. This will be when the real fun begins, when the power of enterprise imaging is unleashed.
Unlock these pixels, and new insights will all but certainly emerge. Quantitative and comparative analysis that will add a new dimension to the textual data mined from electronic medical records. Images may first be analyzed using tags associated with them, tags that point to the signs of disease. These may allow image analysis programs, including deep learning algorithms, to look for patterns in and among images.
The data derived from these images, correlated with those pulled from text, could lead to a new understanding of how imaging can best be utilized — not just by radiologists but by all physicians on a patient care team. Deep learning algorithms might be leveraged to perform data mining. Other DL algorithms might crunch numbers, drawing associations.
Beyond this is a new world of decision guidance in which machines work as partners alongside radiologists and other physicians. Already algorithms are being developed that prioritize the interpretation of images (see “Will Artificial Intelligence Find a Home in PACS?”). Others are being built to support the diagnostic process itself (see “How Intelligent Machines Could Make a Difference in Radiology”).
This future world would be founded on machine intelligence with increasingly intensive interactions between humans and machines. So successful might these machines become that the trick In the future may be to tamp down enthusiasm, to keep human-machine relationships in balance, so people do not become too reliant on them. Physicians will have to decide how much they want to rely on machines — how far is too far.
Since the Industrial Revolution, we have made machines evermore a part of our daily routines. Natural language processing has brought us closer to machines than ever before. Today we ask them to keep track of our schedules, compile lists, play music, answer questions we would otherwise research in books. Now some are being asked to drive cars. As the forerunners of technologies that embed machine intelligence, these vehicles may show us just how far people will allow thinking machines to go.
The earliest adaptations are already appearing as safety nets — braking systems that automatically stop a car if an obstacle is detected; steering systems that bob and weave the car and its passengers out of harm’s way.
Such safety nets will likely make their way into the practice of medicine as parts of clinical decision support systems to keep referring physicians and radiologists on track with best practices that come from data mining. But the real question is what role intelligent machines will play in the mining of data and the establishment of insights that shape the practice of medicine, decisions that ultimately will define what thinking machines do in the future … decisions that are being made right now.
Editor's note: This is the final blog in a four-part series on enterprise imaging. The first blog, “The Impracticality of a Truly Universal Viewer for Enterprise Imaging,” can be found here. The second blog, “Will Enterprise Imaging Save Hippocratic Medicine?” can be found here. The third blog, “Why Workflow Engines Must Work Right,” can be found here.