I'm fucking dying here
Apparently they reverse engineered the reasoning behind an ML model trained to find tuberculosis to see how it figured it out and it used the age of the machine that took the MRIs; the actual MRI was just completely ignored
Apparently they reverse engineered the reasoning behind an ML model trained to find tuberculosis to see how it figured it out and it used the age of the machine that took the MRIs; the actual MRI was just completely ignored
@reiddragon which shows you the importance of grooming your dataset. Likely the training data positives came from one machine, or a set of studies in some time frame and the negatives we largely collected more recently. And the inputs included details of the particular machine the images were captured on, likely with the good intention of controlling for hardware differences.
Anyone who says "just throw it at AI" is blowing smoke. Useful tool, but dependent on competent experiment design.
Follow
@Flux @reiddragon And this is why "yearly MRIs" can be a good idea (if offered for cheap to capture more pop variance, etc etc) despite the "the scans show nothing" naysayers.
