This weekend I immersed myself in IBM’s Watson product, a compendium of arcane AI techniques combined into a platform for a second machine revolution, whose genesis surrounds us today in the form of vast field of literature, a broad expanse of interdisciplinary theories, fully accessible only to those with a deep background in mathematics and computer science.
Watson is fascinating, yet, for me, the coming of this new age is imbued with a sense of inadequacy; a sense that the pudding textured, flabby mess of connections, encased in my skull can’t compete with such breath and sophistication. Will I eventually lose my job to a “machine”? How degrading that would be.
If I feel like this, imagine the reaction of radiologists who have spent the most productive years of their life building pattern recognition skills, understanding anatomy, and matching symptoms to objective findings. It’s been widely reported in the media that they will soon be defunct, as IBM and Merge merge into a unit focused on replacing them with Watson.
Dr Siegel, Director of UM Maryland’s Imaging Research Technologies thinks otherwise.
“I don’t see this technology supplanting physicians or radiologists,” Dr. Siegel said. “It’s a tool that will gather, summarize and analyze information—very similar to the role now performed by our best residents and fellows.”
In short, Dr Siegel sees the role of the radiologist being informed by Computer Aided Diagnosis and Clinical Decision Support Systems, such as those we already have for stenosis detection on CT angiogram, or breast cancer screening, but at a broader scale, with the radiologist being guided to the areas of interest and asked for over-reads.
When we talk about general medicine, IBM’s goals are more lofty, seeking to allow the physician to pose questions, and then applying information sources to form and test hypothesis that are synthesised into confidence-scored recommendations. Watson will draw upon a wide range of unstructured data including “treatment guidelines, electronic medical record data, notes from physicians and nurses, research materials, clinical studies, journal articles, and patient information”.
As an interoperability guy, I have some concerns about the use of unstructured text, and how the complexity of this data will make IBM’s work harder. After all, unstructured text is messy, contains arcane abbreviations, and is built around a context sensitive lexicon. Opining on the HealthCare Blog Po-Hao Chen, MD, said;
medical lingo is anything but natural. A helpful radiology request might read, “75 yo M w/ MM, AAA s/p TEVAR c/b EL on 2/2013 p/w CP r/t back.” A less helpful one: “Unspecified,” Medical lingo has typographical errors, missing punctuations and ambiguous acronyms.
If we, as humans, cannot unambiguously determine the meaning of “MM” in this sentence, how can we describe the algorithm needed for a machine to do so? How will this impact Watson’s confidence scores?
The irony here is that we have just gone through a major effort in structuring medical data in the form of meaningful use, yet it’s not at all clear that the data that’s been collected about the patient, and orders for their procedures, is suitable for Watson’s use. Narrative remains king as physicians are caught in the tension between fully describing the patient condition in code, and getting enough work done to stay viable in a complex system of reimbursements.
To be successful, Watson will need to build a complex view of a patient using a large set of data sources. It’s likely that patient data will be stored outside the institution in Watson’s cloud, and if IBM’s product managers have any common sense at all, it will have occurred to them that if they can get all that data in one place, they will have built the most comprehensive view of that patient anywhere; a valuable prize indeed.
On the flip side, for IBM to efficiently do what is proposed, they are reliant on standards and interoperability to get data from where it resides, and incorporate it into their model. This is awesome news, because it means that even as meaningful use starts to fade away, we have a highly motivated corporate entity with deep pockets, who may be able to lead an industry consortia to success where government failed.
Perhaps this will keep interoperability out of jeopardy.