In theory that sounds valid. However wouldn't you leave the door open for the possibility of NTs having some kind of individual disconnect from their facial expressions versus what they are really feeling, whether involuntary or voluntary? That for whatever reason physiologically while in theory they should react in such a way, they don't necessarily.
My line of thinking parallels issues with the weaknesses of polygraphy. That most people's responses to a polygraph are measurable and subsequently reliable. However some person's responses, for whatever reason are not.
Ah, but that's the difference with microexpressions! Microexpressions are specifically the muscle responses that occur before the conscious processing of what has been said. The body reacts before the mind processes (ever seen something disturbing? I don't mean to choose a heavy example but it's the best I can think of - when I found my roommate after she committed suicide, I looked away and actually said "oh my goodness" before I realized why i was looking away or what was shocking. My brain processed that I was looking at something traumatic, my body reacted, and THEN I processed why. Perhaps another example would be if you've ever been startled, and you involuntarily jump a little even before you've processed what startled you), and that initial pre-processing reaction is what I'm referring to. People can absolutely change their reactions to fit what they want to demonstrate, or what they think they should feel; someone who is actually relieved about a relationship ending can convince themselves to cry, to be sad, because it's the reaction that they think is appropriate, and so forth. The conscious interpretation and expectation can skew the physiological reaction. However, I'm not talking about consciously-affected reactions; I'm talking about the initial, instinctive facial expression, the one that lasts a fraction of a second. Neurotypicals and those lower on the autism spectrum have an instinctive understanding of those expressions that they make instinctively, and they therefore naturally draw them out - a smile, if the person believes they should be happy, may last several minutes, or longer, or a person may put on a fake smile to imitate a response they know would be there if they were truly experiencing the emotion they want to portray. But the science I'm talking about isn't the conscious, deliberate facial expression; those facial expressions exist BECAUSE of the instinctive, scientifically-true microexpressions, but to say that the deliberate facial expression is as real as the instinctive microexpression would be like saying that a potted plant in someone's living room is as much a representation of the natural ecosystem as a rain forest. People can affect science, but that does not eliminate the scientific basis.
As for your comparison to a polygraph, a polygraph is unreliable because it only calculates a handful of physiological responses that would all be present for multiple different emotions; a person who is sexually aroused will experience a quickeed heart rate just as someone who is frightened would. A polygraph doesn't calculate every aspect of the physiological response, so it can't differentiate between emotions that present similarly. I suppose I would compare the polygraph to looking only at, say, 3 of the muscles in the face (out of over 40 that have been identified). A polygraph probably considers pulse, perhaps moisture on the hands... maybe 2 or 3 factors of the entire response. To consider only three muscles of the face might give you a HINT about the actual response is (if you know the eyes are squinting shut, you can narrow down your options - happiness, surprise, shock, etc would not present with narrowed eyes, while anger, hatred, disgust, etc would...) but without knowing exactly which muscles are active in exactly which ways, you can only narrow down, not specifically identify. If a polygraph could measure physiological responses that are specific ONLY to emotions associated with lying (fear, anxiety, etc) it would have the potential to be much more accurate; if it could also calculate this response in the first couple milliseconds after the question is posed, accuracy would further improve as the polygraph would catch the reaction BEFORE full brain processing, rather than when the subject has had the opportunity to process the difference between the emotion they instinctively feel and the emotion they want to portray. However, as the polygraph (at least to my knowledge) is still based on reactions less instantaneous (heart beat, moisture on the palms...none of these is significantly representative within the pre-processing milliseconds, to read the response you have to consider post-processing miliseconds as well - both reactions take more than a millisecond to present measurably), it lacks the accuracy that can come from the tensing of facial muscles - which can be caught on camera then broken down into still images, which can then be used to identify which muscles instinctively tensed before the brain processed what facial expression would be most appropriate.
Wow I'm sorry this is so long, like I said this was something of a special interest for a while...