Two Stations Ahead

A cassette comes open and the tissue sits in the well while the forceps warm. Somewhere between four and fifteen seconds pass before it settles into the mold. Whatever happens in that window is the whole of embedding, and none of it gets written down anywhere.

In 2006 I wrote a handout about this for a seminar, and I wrote it as an infomercial. Bold type, a fake revelation, exclamation points, a promise that the technique would not make anyone rich or better looking. The embedding chapters available then were accurate and joyless, page after page of shaded blobs with a knife edge drawn beside them, and none of that stayed in the head of a tech four hours into a run. Comedy was a delivery mechanism. The jokes have held up better than the acronym I coined to hang them on.

One sentence from that handout was worth keeping, and I nearly buried it under the layout. As embedding is done, one must keep in mind how his/her technique will affect the microtomy and reading of the slides. Phrasing like that dates itself, and the instruction sits where seminar material always sits, in the imperative. Underneath is a claim about reach. What gets decided at the mold lands somewhere else, on another bench, later, in hands that were not there. Everything surrounding that sentence was orientation practice already documented in half a dozen places. The sentence itself was the argument, and it took me two decades to notice I had written it in passing.

Embedding is the only bench station where an irreversible spatial commitment gets made on behalf of two people who are not in the room, about a surface that does not yet exist. Grossing decides what goes in the cassette. Processing decides whether the piece can be cut at all. Neither of those reaches forward the way this one does. Position a fragment and the block face has been chosen, along with the sequence in which structures will appear, the depth at which the last of the specimen disappears, and how much of the tissue a single section can carry. That face is a prediction about a plane nobody has seen yet, made from a wet surface under paraffin, in a few seconds, by someone whose hands are already reaching for the next cassette.

Predictions of that kind are cheap to get right and expensive to get wrong, which is the part rarely counted honestly. Three seconds not spent looking at a fragment can produce a block that goes back to the melter, a face already partly consumed before anything usable comes off, a case pulled from its batch and walked through by hand, and a slide that arrives after the reader has moved on to something else. Skipping the look saves three seconds once. Recovering from it costs several minutes, spread across at least two people, on a day when the count was the reason for hurrying in the first place. Nobody needs a slogan to do that arithmetic. The multiplier is visible to anyone who has stood at a melter with a block they cut themselves.

There is a version of this essay from twenty years ago in which the whole argument ends at a pathologist behind a scope. That vertex has moved. A pathologist working a partial section can read around a gap, tilt the plane in their head, compensate for a fragment sitting proud of the others, and sign the case out without mentioning any of it. Silent correction of that sort is a real skill and it absorbed a great deal of variable embedding for a long time. Scanners do not absorb anything. Tissue sitting at inconsistent depth yields a section where one fragment is complete and its neighbors are grazed, and segmentation treats the grazed pieces as smaller objects rather than as incomplete views of larger ones. An algorithm reports what it measured. Nothing in the output distinguishes a fragment that was small from a fragment that was cut too early.

Pharma sharpens this further. A study block may be the only one from that animal at that timepoint, cut into a design where the section carries a number into a report. Orientation errors there do not generate a recut request. They generate a data point that looks like biology.

None of which changes what actually happens at the mold, because the thinking has never been about technique. Orientation rules are written down and they are not difficult. What is not written down is the question underneath them, which is what this piece of tissue is and what needs to be seen in it. Answer that and the physical decision follows without much deliberation. A tubular structure gets stood on end because someone recognized a lumen. Fragments get lined up in a single row facing the same way because someone noticed they came from one surface. Skin gets placed so the keratin meets the blade last because someone thought about which layer would compress the ones beneath it. Skip the identification and everything downstream is a coin toss executed with excellent hand skills.

Newer techs sometimes hear that as a demand to slow down, and it is not. Recognition speeds up considerably with practice, to the point where the experienced hand appears to be doing nothing at all before placing tissue. Watch closely and there is a pause of well under a second while the piece gets read, and the placement that follows is not deliberated because it was already settled by the reading. Asking the grosser closes the gap faster than any amount of independent staring, and there is no cost to asking beyond a moment of feeling new.

Embedding is unusual among bench decisions in that it leaves a durable physical record. The block outlasts the shift, the tech, and frequently the case. Fragment depth, paraffin margin, the angle at which tissue meets the blade, all of it sits in a drawer months later, fully legible to anyone who faces the thing and takes a few cuts. So the recognition problem here is stranger than simple invisibility. Evidence of the decision exists, and it is readable, and it reaches only the audience with no standing to comment on it. A pathologist receives a section rather than a block. An algorithm receives a mask. Neither one ever handles the object that holds the record. Judgment at the mold gets read sideways, by other techs, at the microtome, usually in the form of quiet gratitude or quiet swearing, and never in any form that travels upward to the people deciding what the work is worth. Meanwhile the count survives into every metric a lab keeps, and blocks per hour is the one dimension of this that carries no information about whether any of it was done well.

Good embedding does not guarantee anything. Fixation still fails, processors still run short, blades still chatter, and antigens still vanish for reasons nobody traces. What it does is remove one source of failure from a list that stays long regardless. That is a smaller promise than the one I made in 2006, and closer to what the bench actually delivers.

Thin Sections is written by Jerry Clarin, HT (ASCP). The views expressed are his own and do not represent any employer.

The 2006 seminar handout, six pages, as distributed: