So I have to ask, of you PRACTICING radiologists (not pre-med pessimists), do you think that the concern of computers taking over the field of diagnostic radiology is a justifiable threat in the foreseeable future (15-20 years)?
When it comes to the future of radiology, I sleep very soundly at night. Most clinicians I know have a very, very poor understanding of how a radiologist functions, so I always find it amusing to hear them wax poetic about radiology's future. If someone in medicine is so clueless, then a pre-med or med student's perspective is that much more likely to be way off.
There are a lot of wonderful and mind-boggling things that technology and computer science can accomplish, but pattern recognition + weighted judgment rivaling a human isn't going to be one of them, probably not in my lifetime at least.
Our human brains are just wired exceptionally well for this task. It's a combination of tremendous processing power, ~30 years of software refinement through living and navigating in the world 24 hours a day, and ~10 years of medical knowledge and training that allows humans to make judgment decisions on imperfect studies that often don't show the textbook features of a pathology. CAD for breast is pretty crude still, mostly looking for microcalcifications and other easy-to spot anomalies. And my prediction is that while computing power/speed will increase by about 300% in the next five years, breast CAD will only improve in accuracy by perhaps 10-20%. The computing power required to equal what rads do naturally (well, naturally + 10 years of medical training) just absolutely dwarfs what will be available for a couple of decades.
Let's put it another way... the edge detection processing done in our retinas alone far exceeds what the best algorithms running on the fastest computers are capable of today.
Our brains may not be wired to add huge numbers quickly, but we are wired to see pretty darned well.
Maybe internists should worry first. It is easier to teach a computer how to follow patient management algorithms, by far, than to teach it how to read even a simple X-Ray. Which is to say internists should worry about computers taking over their jobs (or at least making all the treatment and management decisions) before radiologists should worry about computers taking over their's. High blood sugar? Okay, a computer can figure that out and then spit out a drug choice. You'd just need thorough nurses to enter all the data in and physical exam findings (not that physical exams actually guide much anymore).
I'm speaking tongue in cheek; I don't think internists need to worry about computers taking over their jobs. Just saying it could happen to them much earlier than it would happen to us.
Radiology is as much art as science, and context plays a large roll. It's not chess or math where the rules are hard and fast and where "right is right". Thus computers can't be expected to master it anytime soon.
Examples: Identical ultrasound findings can mean diametrically opposed things, but with context of comparison X-Ray it can be easily distinguished (emphysematous cholecystitis is a mimic for porcelain gallbladder on ultrasound, for example, but easily distinguishable with correlation). Identical findings can be suggestive of different things depending on clinical history: free air in a seventy-year old with bowel obstruction can relate to perforation, but free air in that same seventy-year old can be normal status post surgery. Free fluid may be normal in a teen girl but is rarely normal in a boy. White matter disease is chronic small vessel disease if you are seventy but may be Lyme, multiple sclerosis or guillain barre in a thirty year old. Etc..