Slynnpack Equine musculoskeletal imaging

Research and evidence

What is actually established.

We separate three things carefully: published results by other people, public market facts, and our own claims. The third list is currently empty.

§01Published, by others

Small data was enough.

A peer-reviewed study graded sesamoiditis on equine radiographs using a two-stage detector, from a dataset most machine learning practitioners would call far too small.

The radiographs came from an equine practice in central Kentucky, and a veterinarian from that practice appears among the authors. We have no connection to this work. We cite it because it establishes the floor for what this problem requires.

Citation

Deep learning model shows promise for detecting and grading sesamoiditis in horse radiographs

Guo L, Yu X, Thair A, Rideout A, Collins A, Wang Z J, Hore M. American Journal of Veterinary Research, 2023.

Radiographs
255
Views used
DLPMO and DMPLO
Classes
Normal, mild, moderate
Architecture
cascaded RetinaNet with a self-attention classification head
Accuracy
92.7%
Mean average precision
81.8%
DOI
10.2460/ajvr.23.07.0173

§02Public market facts

The context, from published sources.

  • Minimum radiographic views required per yearling by a major North American sale repository 3638 at racing age
  • Days before selling by which the study must be filed 4
  • Maximum age of the radiographs when filed 21days
  • Studies reviewed by the auction house itself 0
  • Radiographs behind the published sesamoiditis result 255

Repository requirements are drawn from the published terms and digital requirements of a major North American yearling sale. Sales differ and terms change between renewals. Always work from the conditions published for the sale you are entering.

§03Our own results

None yet.

The company was formed in September 2026. We have not trained a model on a partner archive, because we do not yet have one.

When we do publish results, they will carry the dataset size, the class balance, the confidence intervals, the panel agreement statistics and the failure cases. Reporting accuracy alone, on a favourable test split, is how this field earns distrust.

Until then, please read anything on this site describing the product as a statement of intent rather than of capability.

§04Regulatory position

Stated plainly.

Software intended for diagnosis in animals is treated differently from software intended for diagnosis in people. The premarket notification requirement that governs human medical devices is written to reach devices intended for human use, and veterinary products are not subject to that clearance process.

That is a genuine difference in the regulatory path, and it is one reason this work is tractable for a small company. It is not a claim of safety, of quality, or of anyone's approval.

No product described on this site has been reviewed, cleared or approved by any regulatory authority, and none is offered as a substitute for examination by a licensed veterinarian.