You're making some assumptions here that are not entirely warranted. This is not to say that your assumptions are wrong, but they could be right or wrong, and treating them as foregone conclusions both hurts your credibility when debating this issue and casts an unpleasant light on those of us who agree with you but show less tinfoil paranoia when doing so.

You're assuming this workflow on a NoS with ATR:

1) NoS emits MMW energy
2) MMW energy reflects off suspect, er, targeted individual
3) NoS receives reflected energy
4) NoS composes a digital image from the scan returns
5) ATR software analyses the digital image for pixels of a color inconsistent with human skin
6) ATR software lights up the appropriate portions of the stick figure indicator

However, I believe that there is a different workflow:

1) NoS emits MMW energy
2) MMW energy reflects off suspect, er, targeted individual
3) NoS receives reflected energy
4) NoS passes the raw scan data to the ATR software without composing an image from them
5) ATR software analyses the raw scan data for energy return signatures inconsistent with human skin
6) ATR software lights up the appropriate portions of the stick figure indicator

Certainly, given the way that TSA works, the idea of the ATR software analyzing a digital image rather than the raw scan data fits with the bureaucratic insanity of the agency. However, we should remember that the NoS and the ATR software were designed by engineers who work for private contractors, not personally by John Pistole or by any of the 50,000 middle management dunsels who comprise the bulk of the TSA.

So, while the NoS almost certainly still contains the legacy software necessary to convert raw scan data into a naked picture, I don't believe that the NoS currently uses that software, instead routing the raw scan data directly to the ATR software add-on for analysis.

TSA has always stated that the NoS never stored or transmitted the images. But I always sensed some intentional misrepresentation there; they never said anything about storing or transmitting the raw scan data. And it's possible that the raw scan data could be stored, perhaps even in non-volitile memory, and could be used to create a finished image at a later time.

We don't know how much volatile memory the NoS is equipped with. It may have enough of a volatile buffer to hold multiple images, or perhaps multiple sets of raw scan data, or both. My 15.1 megapixel Canon 50D has a volatile buffer which can hold up to 10 images while it waits for the memory card to catch up to it, so I wouldn't be surprised if the average NoS has buffer memory sufficient to hold at least several sets of raw scan data or several finished images, or both. And there is no guarantee that the buffer is completely volatile memory; it couls be a flash drive designed to be overwritten with each new scan, theoretically leaving the last scan permanently in place till the next scan occurs, or perhaps leaving recoverable imprints of the data like a deleted file.

The point is, while all of this is theoretically possible, saying that "it IS!" definitively is just as foolish as saying that "it ISN'T!" definitively.