Reading about AI algorithms for FDA

[sbmthakur]

https://drive.google.com/file/d/11JqWMGLnd5qmIWbNxjh3EJlDDqbS4L40/view

[sbmthakur]

The researchers have identified that ML algorithms in radiographic detection did not work as expected. For instance, they quote a report (by DeGrave and colleagues) in which the algorithm learned "shortcuts" in image acquisition instead of following a pathological process.

They conclude the paper by mentioning that clinical applications of FDA-regulated ML algorithms is difficult to justify due to the non-inference of generalizability and the presence of bias.