A.I Versus M.D. – Part 2

| July 16, 2017

In April I wrote a post entitled A.I. Versus M.D. – Either Way – Pathologists Do Not Diagnose Cancer – Part 1. This was followed by a guest post by David West, CEO, Proscia, entitled AI vs MD – What it Means for Pathology in May where David, among many issues points out the use of “intelligent” algorithms and what his company sees as the future of AI in pathology.

Aside from my rants about lack of mention of pathologists diagnosing disease in the original New Yorker article, I mentioned I would share thoughts on man versus machine in a second part.  What I didn’t fully appreciate when part 1 was posted, prior to FDA clearance of the first clinical system for primary diagnosis by Philips, is the amount of interest AI has captured in the digital pathology market. As of the Digital Pathology Congress last week in Chicago, there is clearly more interest in digital pathology. Certainly, for business cases using primary diagnosis but also using repositories of images for AI and developing algorithms for clinical use.

So, here goes A.I. Versus M.D. – Part 2

terminator1As a medical student in the early 90’s I was part of a group that piloted a software application that was attempting to help medical students, and eventually physicians arrive at differential diagnoses more accurately and determine best next steps in terms of where to focus physical examination findings and what radiology and/or laboratory tests to order. The Matrix was several years away. The Terminator was released a decade earlier. We watched as John Connor and his resistance forces defeated the machines directed by artificial intelligence Skynet.

TheMatrixThe idea was if a patient presented with “fever and cough”, the application would generate a differential of 47 possibilities, ranking them in order of most likely that could be further refined by knowing the patient’s age, sex, social history, family history and then guide us where to concentrate our review of systems and physical examination to refine this further. So, you might put in that the symptoms were acute, the patient was a non-smoker, without recent travel, no known sick contacts, had rales on physical examination and then the application would tell you what might be appropriate tests to order (i.e. chest x-ray, CBC with differential, sputum gram stain and culture, etc…).

It turned out that the application was no better than a second-year medical student armed with a $5.99 pocket book of differential diagnosis and appropriate work-ups. And the application was not portable as there were no smart phones or mini IPads with this application loaded to use on the wards. We were still using one-way pagers and land lines to return those pages.

I do not believe that application was ever brought to market. Perhaps other companies or improvements on this have been commercialized but I recognized something early on that still holds true today I think, for the most part.

If it is hard for a human to do it, it is hard for a human to program a computer to do it. If it wouldn’t replace physicians, perhaps it could help them provide more cost-effective care by generating time savings, directed reviews and efficiencies in gathering information or identify key data points.

Fast forward 25 years, self-driving automobiles in pilot and more large investments into Artificial Intelligence, the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages, is an active area of research and development.

The economics of this long-term are interesting if you consider long-term what jobs, tasks, skill sets can/will be replaced and the impact to society if you believe what is being written about The Real Threat of Artificial Intelligence.

For those of us in pathology and healthcare, of course our attention turns to what this means for our practices and our patients. On one hand, most of us would welcome levels of automation and/or computation that can eliminate tedious and repetitive tasks without compromising accuracy, precision and consistency.

It comes down to what can computation do and what is the role of cognition. While not necessarily AI, automation and computations have automated clinical pathology. Serum chemistries, toxicology, urinalysis, now blood banking and microbiology have all benefited from computer technology replacing what were once time consuming manual tests with inherent variability that comes from any “testing system” that we try to control as much as possible.

Now attention has turned to anatomic pathology. The question is not when will AI be used, but how it will be used?

There is little doubt the technology will enable pathologists for rare event detection, i.e. acid fast smears, malaria smears, isolated micro metastasis, focal invasion, rare lymphvascular space invasion, spatial relationships between lymphocytes and tumor cells and more. Needle in a haystack approaches that human cognition will benefit from computer computation.

One will not replace the other, but other enable the other. Now by definition, these systems have the capability to learn over time, acquire additional knowledge from experience, as do humans, to refine their approaches and identify data that was previously not within their bits of bytes of programming. There is no substitute for experience.

But beyond that, I think pattern recognition, morphologic interpretation, the study and diagnosis of disease that pathologists do daily is only part of the formula to the practice of pathology. That is one source of data but not the only source to arrive at a correct diagnosis. Often, we need sources of data from multiple disparate sources to formulate a clinicopathologic diagnosis.

Perhaps this is not far of either if AI can also read and mine electronic medical records, clinical laboratory tests and acquire clinical experience and knowledge and read the current literature to refine their approaches to diagnoses.

But I think for now we are safe from rebelling against matrixes and machines.

Tags: , , , , , , , ,

Category: Advocacy, Anatomic Pathology, Business, Clinical Laboratories, Clinical Pathology, Cloud Computing, Conferences, Current Affairs, Data Management, Device Manufacturers, Digital Pathology News, Education, General Healthcare News, Image Analysis, Informatics, Laboratory Informatics, Medical Research, Pathology News, Science

Comments are closed.