The Economist has a revealing article about a new radiology system for detecting breast cancer from Duke University Medical Center. (It's behind a subscription wall but you can read it if you suffer through an ad, natch.)
Dr. Georgia Tourassi and her colleagues have developed a mammography system that analyzes radiology images for tumors by comparing them to an archive of images of known, diagnosed tumors, rather than by brute-force feature abstraction. It also acts more like "an intelligent colleague than a black box" – offering users rationales for its results based on comparable images. As more images of established tumors are added to the database over time, the system has more basis for comparison.
The system attempts to address two problems –
- the frightening statistic that radiologists relying solely on their own eyes and judgment can miss up to 30% of tumors – two-thirds of which seem visible through imaging retrospect
- that computer-aided detection systems can help, but deliver lots of false positives and offer radiologists an unsatisfying and opaque experience. Without an explanation for an alert, radiologists often "are reluctant to accept a diagnosis at odds with what they think their eyes are telling them." Which is totally reasonable. Who accepts information from a computer that does not accord with one's own judgment and perception of the outside world without some evidence offered?*
What's interesting here is both technological and social. If you believe the story, previous computer-assisted-detection systems have apparently failed to win acceptance because they function as mini-gods, not colleagues. That is, they demand belief instead of providing proof.
Which is kind of funny, in a macabre way – that's how many patients feel their doctors treat them.
As my perceptive friend Elizabeth pointed out, the most deeply felt knowledge struggle in healthcare right now – again, especially in cancer – is often between patients and healthcare professionals. While providing radiologists with a more trusted resource in examining images seems to be a good thing, there's still plenty of improvement in the patient-practitioner relationship. Where do we fit "nothing about me, without me" in this picture?
*Well, not these people. But that's a whole different issue.
