Algorithmic Limitations

Note: the first example is more like a constraint than a limitation.

Summary

Algorithm limitations

When the FDA talks about limitations, they want to know more about scenarios where your algorithm is not safe and effective to use. In other words, they want to know where our algorithm will fail.

Computational limitations

If your algorithm needs to work in an emergency workflow, you need to consider computational limitations and inform the FDA that the algorithm does not achieve fast performance in the absence of certain types of computational infrastructure. This would let your end consumers know if the device is right for them.

Medical device reporting

After your algorithm is cleared by the FDA and released,
the FDA has a system called Medical Device Reporting to continuously monitor. Any time one of your end-users discovers a malfunction in your software, they report this back to you, the manufacturer, and you are required to report it back to the FDA. Depending on the severity of the malfunction, and whether or not it is life-threatening, the FDA will either completely recall your device or require you to update its labeling and explicitly state new limitations that have been encountered.