Some more detail (from https://www.fda.gov/MedicalDevices/DeviceRegulationandGuidan...): What is Substantial Equivalence

A 510(k) requires demonstration of substantial equivalence to another legally U.S. marketed device. Substantial equivalence means that the new device is at least as safe and effective as the predicate.

A device is substantially equivalent if, in comparison to a predicate it:

has the same intended use as the predicate; and has the same technological characteristics as the predicate; or has the same intended use as the predicate; and has different technological characteristics and does not raise different questions of safety and effectiveness; and the information submitted to FDA demonstrates that the device is at least as safe and effective as the legally marketed device. A claim of substantial equivalence does not mean the new and predicate devices must be identical. Substantial equivalence is established with respect to intended use, design, energy used or delivered, materials, chemical composition, manufacturing process, performance, safety, effectiveness, labeling, biocompatibility, standards, and other characteristics, as applicable.

Only if your intended use also remains the same (i.e. for detecting wrist fractures).

No, it's contingent on what exactly the classifier does and how independent of third party clinical judgement it is. The more hands off the clinician is, the more dangerous a false negative is, and the more time sensitive the result is, the more likely your diagnostic is going to be class 3.

I would naively expect the bulk of new low risk ML product submissions to be de Novo submissions due to lack of real precedent.