How it works
Ingest, audit, read, carry
Four stages. No model names, no thresholds and no data sources, because those are not ours to put on a public page — but nothing here is vague about what the product actually does.
Stage 01
Ingest
Files arrive and become a study for one horse. Personal information is stripped on the way in, each file is checked to be a radiograph at all, and provenance is recorded for every one of them.
A repository turns up as a folder. Before anyone can work with it, somebody has to open every file and work out what is actually in there.
EquiRay does that on arrival. A film carrying a client's name does not carry that name into our system, because the personal information is removed at ingest rather than later. That stripping is live, and it is pinned byte for byte against a second implementation so the two cannot drift apart.
Provenance is kept for every file: where it came from, when it arrived, and what was done to it.
Stage 02
Audit
The set is compared against the repository standard. Each image is given an anatomical view and an image quality score, and anything the standard expects but the folder does not contain is named.
Every image is classified by anatomical view, and scored for whether it is good enough to work with.
The set is then audited against the standard. What is here, what is missing, and what is too poor to be worth opening.
A missing view is treated as a risk in its own right, because it is one. A gap in the set is a real thing to know about a horse, and it is the kind of thing that quietly disappears when a folder is handed on unopened.
Stage 03
Read
The veterinarian works through the study in a reading room built for it, measures what they need to measure, and records findings against the study, with their name on them.
The reading room organises the work, keeps the study together, gives you the measurement tools, and gives the findings somewhere structured to live.
A model locates the landmarks on the film. The code computes the measurement from those points, and you can drag any point and watch the number change. No number is read off the image by the model, so every measurement is one you can audit rather than one you have to trust.
Measurements carry through to the next sitting, so the same structure measured twice can be put side by side, or a whole series laid out in the order the films were taken with the interval between each and every gap in the chain named.
What the product refuses to automate is the finding. It comes from the veterinarian, it is recorded as theirs, and it stays theirs.
The strongest verdict the software will ever reach on its own is that something needs review. A veterinarian accepts, and a veterinarian decides everything after that.
Stage 04
Carry
The findings and the veterinarian's own written report feed the buyer's risk read, so what the vet said travels with the horse instead of sitting in an inbox.
The write-up reader takes the veterinarian's own written report, in prose, and pulls out what was said so the buyer's risk read can use it.
This is the one people mishear, so it is worth being exact about. It works on the words the veterinarian wrote. It does not work on the pictures.
A buyer is handed a repository and a written report and half understands both. They do not know which of eleven findings is the one that matters for the horse they want in work next winter. Carrying the vet's own words through is how that gets better, and it does not require a machine to have an opinion about an image.
Where the line sits
The strongest verdict the software will ever reach on its own is that something needs review. A veterinarian accepts, and a veterinarian decides everything after that.
EquiRay measures. It does not diagnose, and it will not tell you what a change means. A finding nobody checked is worse than no finding, because somebody will act on it. So the software organises, checks and carries, and the veterinarian judges.
Early access
We would rather build this with a handful of vets who will tell us plainly what is wrong with it
EquiRay is in development. If that is of interest, tell us what you image most and how the repository reaches you today.