Ann: ResApp Receives First Institutional Review Board Approval-RAP.AX, page-125

The algorithm will continue to get better as long as it continues to be trained. For that you need outcomes to compare to predictions. That's easy enough doing trials with relatively low numbers. But once you're dealing outside of trials with patients and doctors who have no commitment for follow-up your flow of training data will dry up. You can't keep training on the old data or you just "over-train" the algorithm, which produces worse accuracy when applied to the general population.

Also quite typical of training learning algorithms is that you get a rapid convergence toward the final accuracy followed by refinement producing small, gradual improvements.

We could be getting close to "as good as it gets". The purpose of the US trials is to demonstrate our accuracy. It's unlikely that we will develop better accuracy, although we could produce greater statistical confidence.