Is diagnosis like facial recognition?

Mark Graber Mark.Graber at IMPROVEDIAGNOSIS.ORG
Fri Jul 27 23:44:23 UTC 2018

Amazon was criticized publicly today because its facial recognition software had an apparent error rate of 5%, misidentifying several members of congress as felons.   In its own defense, Amazon said that the trial was done by setting the algorithm’s ‘confidence threshold’ to 80%, and that for important matching a much higher threshold, maybe 95% or more, would be more appropriate.

It will be interesting to see if congressfolk understand sensitivity\specificity tradeoffs, but the article raises interesting questions for us in diagnosis, given that so many (most ?) of our diagnoses are made by pattern matching, as Dr Jain so often reminds us ;<).  Isn’t this analogous to the facial recognition problem?  So what level of specificity should we be looking for before we label someone as having a particular diagnosis?  Each of us probably has our own confidence threshold before we’re comfortable assigning a label to someone’s problem, and I just wonder what error rate that equates to.


Mark L Graber MD FACP
President, SIDM
Senior Fellow, RTI International
Professor Emeritus, Stony Brook University

To unsubscribe from the IMPROVEDX:
or click the following link: IMPROVEDX at LIST.IMPROVEDIAGNOSIS.ORG


For additional information and subscription commands, visit:

http://LIST.IMPROVEDIAGNOSIS.ORG/ (with your password)

Visit the searchable archives or adjust your subscription at:

Moderator: David Meyers, Board Member, Society to Improve Diagnosis in Medicine

To unsubscribe from the IMPROVEDX list, click the following link:<br>
<a href="" target="_blank"></a>

HTML Version:
URL: <../attachments/20180727/b13defe6/attachment.html>

More information about the Test mailing list