We talk a lot about AI in this newsletter, but this week it's another buzzy tech acronym getting its moment: Augmented Reality (AR). A new proof-of-concept study in the American Journal of Veterinary Research found that augmented reality glasses, the kind that overlay digital images onto what you're actually looking at, showed promise for improving spatial accuracy during complex canine head and neck surgeries, without slowing surgeons down.

The setup: researchers at UC Davis had 22 veterinarians use AR eyeglasses to overlay a holographic image of a dog's head directly onto a simulated surgical field. The idea is that for procedures like tumor removal, where precisely identifying and avoiding healthy tissue is the whole ballgame, having 3D diagnostic imaging projected right onto the patient rather than glanced at on a separate monitor could meaningfully sharpen accuracy.

This is early stage, proof-of-concept work, not something you'll see in a surgical suite next month, and the researchers themselves note the next step is figuring out whether it translates to actual clinical settings. But it's a good reminder that the AI conversation dominating the profession right now isn't the only frontier worth watching. Surgical visualization tech has been creeping into human medicine for a few years, and as with most innovations, veterinary medicine may not be far behind.

When it comes to guardrails around technologies, Europe is usually further ahead than North America. The same seems to be true for Artificial Intelligence in veterinary medicine. The Royal College of Veterinary Surgeons has published draft documentation on what AI tool providers should be providing to veterinarians by way of documentation. VIN News covered this development this week. The proposal asks a lot more of AI tool providers than is currently available. It lists 23 items, over half of which are not provided by a majority of diagnostic AI tools in practice. It is great to see these developments moving forward and I know that organizations in North America are working towards similar guidance.

Looking to the future of AI, an interesting pre-print was published on arxiv this month. Zhang et al describe APOLLO, a multimodal temporal foundation model trained on more than three decades of electronic health record data from Mass General Brigham. This training encompasses a whopping 7.2 million patients, 25 billion clinical events, 28 data modalities, and 17 hospitals or affiliated clinics. The goal is to create a “virtual patient” representation that combines structured data, clinical text, diagnostic reports, images, medications, diagnoses, labs, and temporal history into a single embedding. The authors then test these embeddings across 322 downstream tasks, including disease onset, progression, treatment response, adverse events, hospital operations, and patient retrieval.

The strength of the work is that it points toward a future where AI is not just a point solution for one image, one lab value, or one diagnosis. It is trying to model the patient journey. Clinical reasoning is temporal and multimodal. That is, a radiograph, a lab value, a medication, and a prior diagnosis mean more together than they do separately.

But the paper should not be mistaken for proof that this future is already here, or that it will generalize easily. The data come from one very large, highly resourced human health system. That means the model may learn local practice patterns, documentation habits, coding behaviour, referral bias, and population-specific signals. Strong internal performance does not necessarily mean the same model would work in community clinics, different countries, different EHR systems, or lower-data environments.

This is especially relevant to veterinary medicine. Vet med has fragmented records, inconsistent coding, variable imaging metadata, fewer large linked datasets, and far more species, breeds, and body-size variation. The future probably does look like patient-level multimodal AI, but veterinary versions will need purpose-built data infrastructure before we can expect APOLLO-like models to perform reliably across real-world practice. At the same time we are many steps away from these systems being proven help patient outcomes, rather than just look good ex vivo or on paper.

-RBA

Hill's Pet Nutrition has entered the fresh food category for the first time, launching Science Diet Single Protein Dog Food Rolls at pet specialty retailers this month. The steam-cooked, single-protein recipes (chicken and brown rice, beef and carrot, lamb and pumpkin) are formulated to be a complete meal. It's a notable move for a brand built on kibble and prescription diets to plant a flag in the refrigerated aisle.

If your clients are anything like ours, they’re almost certainly going to ask about fresh food (since they do constantly). With a big name behind this new product, it’s worth reminding them of both sides of the equation. Fresh food can work for some dogs: higher digestibility in some cases, and fewer processing byproducts. But the case for caution is real. Fresh and refrigerated diets carry a shorter shelf life and stricter storage requirements than kibble, home-prepared and boutique fresh diets have a spottier record on complete-and-balanced formulation than major manufacturers, and the fresh category overall has less long-term feeding trial data behind it than traditional dry and canned lines.

Hill's entry doesn't resolve that debate, but it does put a major, veterinarian-facing brand's name behind a category that's mostly been the domain of smaller, DTC-style companies.

Here are some of the other stories that caught our eye and we're following this week from around the veterinary world and animal kingdom: