The first thing to establish about US20260203900A1 — published on July 16, 2026 and assigned to Stryker Corporation — is that its first operative claim is not claim 1. Claim 1 is canceled. The independent method claim that carries into publication is claim 2, and it is claim 2, rather than the title or the abstract, that defines the scope Stryker is currently pursuing. This is a pending application, an A1 publication: it has been published, not granted, and nothing in it is enforceable today.

Claim 2 is a method claim, and it recites four steps in sequence. First, receiving a fluorescence image of the tissue of a subject. Second, providing that fluorescence image to a generator of a trained generative adversarial model. Third, obtaining from that generator a simulated white-light image depicting the predicted future state of the tissue. Fourth, displaying that simulated image. The claim limitation that matters here is the second one: the recited component is the generator, specifically. Not the discriminator, not the adversarial pair as an undifferentiated whole, and not a classifier. A system that scored perfusion imagery for necrosis risk without passing it through a generator half of a GAN would sit outside the literal language of claim 2.

The second limitation that does real work is the output. Claim 2 does not claim a segmentation mask, a probability, or a risk score. It claims a simulated white-light image — a synthesized picture that looks like an ordinary visible-light photograph of the tissue, but which depicts a state the tissue has not yet reached. The claimed method is, in effect, image-to-image translation across time: intraoperative fluorescence in, a plausible postoperative photograph out. That framing is what separates this application from the large existing body of computer-aided-diagnosis claims that terminate in a label rather than in a picture.

What the dependent claims narrow to

The dependent claims supply the definitions that claim 2 leaves open. The phrase “predicted future state” is otherwise abstract; claim 3 gives it content by enumerating the outcomes contemplated.

The method of claim 2, wherein the future state of the tissue comprises necrosis, delayed healing, healing, or any combination thereof.— Claim 3, US20260203900A1

Read against claim 3, the method is directed at surgical viability prediction rather than at diagnosis in the general sense. Claims 8 through 10 narrow the input in the same practical direction: the fluorescence image is a NIR image (claim 8), an intraoperative perfusion image (claim 9), or a frame drawn from an intraoperative perfusion video (claim 10). Claim 11 supplies the tissue types, listing breast tissue, burnt tissue, chronic wound tissue, acute wound tissue, and skin transplants — a set that maps onto reconstructive, burn, and wound-care procedures where flap and graft survival is the clinical question and where indocyanine-green perfusion imaging is already in theatre.

Claims 12 through 19 cover the training regimes, and they are drafted to span both halves of the image-translation literature. Claim 12 recites training on a plurality of image pairs, and claim 13 defines what a pair is: a fluorescence image of a particular tissue during an operation and a white-light image of the same tissue after the operation. Claim 14 names pix2pix. Claims 15 through 17 cover the unpaired alternative — unpaired image data, drawn in claim 16 from a set of intraoperative fluorescence images and post-operation white-light images, and in claim 17 collected across a plurality of patients — with claim 18 naming CycleGAN. Claim 19 generalizes to first and second imaging modalities. Naming specific architectures in dependents while keeping the independent claim architecture-agnostic is a conventional drafting posture, and it is worth noting the dependents cover both the paired and unpaired training paths rather than committing to one.

Claims 20 and 21 mirror claim 2 in the two standard alternative statutory forms: claim 20 as a system claim reciting one or more processors, memories, and programs, and claim 21 as a non-transitory computer-readable storage medium claim. The recited steps are identical across all three. There are, therefore, three independent claims in the published set — 2, 20 and 21 — and they rise and fall on the same four limitations.

Two features of the published text deserve flagging, because they sit inside all three independent claims and are reproduced identically in each. The fourth step is recited as “displaying, on the display” — but neither claim 2 nor the abstract ever introduces a display. There is no earlier antecedent for the definite article to refer back to. Separately, the model is recited with the parenthetical abbreviation GAN, while the word “network” that the abbreviation expands to appears nowhere in the recitation, leaving the adjective without its noun. Both are the ordinary sort of informality that antecedent-basis and clarity practice addresses during prosecution, and both are reasons the independent claims here should be read alongside the file wrapper rather than quoted as settled scope. They are reproduced above as published; they are not typographical errors introduced in this account.

Classification and landscape position

The application is classified under G06T 7/0012, the image-analysis class for biomedical imaging, and under A61B 5/7267 and A61B 5/7275 — the diagnostic-signal-processing subclasses covering classification by machine learning and the presentation of a probable diagnosis. G16H 50/20 adds the health-informatics class for computer-aided diagnosis. Three G06T 2207 indexing codes complete the picture, tagging the record for fluorescence imagery, neural-network processing, and biological-tissue subject matter. That combination places the application at the intersection of surgical imaging hardware and applied deep learning rather than in either alone, which is consistent with where the recited method actually operates.

The application does not stand alone in Stryker's recent publication activity. Two applications published the week prior sit in the same imaging-AI cluster: US20260195867A1, directed to medical image enhancement using machine learning, and US20260195933A1, directed to real-time processing of medical imaging data using an external processing device — the latter addressing the compute-placement problem that a generator running against live intraoperative video would present. US20260199018A1, on visually guiding bone removal during a joint procedure, published the same day as the hero application. Adjacent hardware records round out the visualization estate: US20260206113A1 on maximizing surgical-light output and US20260185929A1 on characterizing fluids flowing through a conduit using optical emitters and detectors. Aggregated, the cohort reads as a set of filings around the surgical visualization stack rather than a single isolated software claim — though a cohort of this size supports an observation about filing direction, not a conclusion about portfolio strength.

For anyone tracking this record, the operative questions are prosecution questions. Whether the generator limitation of claim 2 survives examination in its published form, whether the display antecedent is corrected by amendment, and whether the architecture-specific dependents at claims 14 and 18 end up carrying the allowable subject matter are all matters for the file history, not for the publication. As of today the claim set described here is the claim set as filed and published — a snapshot of what Stryker asked for, not of what it will receive.