There’s a series of US Department of State documents from the 1960s called the International Boundary Studies. Most can be found on the web with a Google search, if one knows the keywords. A copy of the one describing the Thailand-Myanmar border can be found at https://fall.fsulawrc.com/collection/LimitsinSeas/IBS063.pdf . Mostly, the border will follow rivers or mountain ridges, so if the terrain is clearly visible enough in the imagery one can draw a reasonably accurate representation by following the zigzagging spines of the mountains. The level of detail, though, probably isn’t going to be much better than that of the LSIB data (the source of the latest import). We’re talking inaccuracies of mostly less than 100 m; this will require visible detailed features such as fences (as with the Malaysian border mentioned by Bernhard Hiller) or, as AlaskaDave noted in the OP, roads. (I wouldn’t actually expect to find entire villages on the wrong side with this data source, though.)
Personally, I’ve only manually adjusted borders where the visible features are very obvious, e.g. there’s a fence, or a pair of roads that clearly belong on different sides. A single road seems rather difficult to gauge—does it run exactly alongside the border, or how far away is it (or even whose side does it belong to)? In this case, the on-the-ground information would indicate that this road should be on the Thai side, so I guess it should be safe to adjust accordingly.
PS If anyone feels like it they could try checking the LSIB (Large Scale International Boundaries) to see if the latest version is more updated/accurate in the relevant areas (though I don’t think it’s likely).
Edit: Oh, just re-read the OP and saw that the suggestion was that the road exactly corresponds to the boundary. I’m not so sure about this; seems like a rather strange arrangement.