• Fragmented data + tight telehealth rules. Evidence exists, but translating it into compliant, usable workflows is inconsistent.
• Build VCPR-safe, explainable tools embedded in the record. State-aware guardrails, imaging quality checks, overlays tied to familiar measures, and model fact sheets reduce risk, speed reads, and remove clicks that block adoption today.
• See between visits with validated wearables and simple playbooks. Short sprints, exception ladders, and owner education trigger timely calls, therapy adjustments, and visits.
• Turn notes and pixels into decisions with Natural Language Processing (NLP) and imaging support. Structured snapshots, stewardship dashboards, review screens, and governance improve triage, prescribing, follow-ups, and accountability.

Veterinary care produces rich signals that rarely reside in one place. A thoracic radiograph is excellent at one point in time. The owner’s video captures a hacking episode the clinic may never see. The record documents impressions, medications, and plans in prose. The home shows daily patterns that reveal itch, sleep, play, and stiffness. Today, these pieces are often disconnected, forcing clinicians to rely on limited snapshots and memory.⁶

Telehealth expanded access but codified guardrails. In the United States, most telemedicine requires an existing VCPR.¹ The American Veterinary Medical Association (AVMA) provides guidance as a starting point for clinics.² In the United Kingdom, updated “under care” guidance clarifies how remote interactions relate to prescribing and responsibility.³ When tools ignore these constraints, they create risk for clinics and confusion for clients. The lesson is to align with rules, then design for value at the bedside.⁴

The technical frontier is promising. A comprehensive review outlines how AI is entering veterinary diagnostic imaging with growing maturity.⁷ Deep learning detects canine cardiomegaly on thoracic radiographs and identifies left atrial enlargement with strong discrimination.⁸ A separate pilot on thoracic radiographs also demonstrates accurate detection of left atrial enlargement in clinical settings.⁹ Image quality is a recurring pitfall, which is why quality detection for underexposure or rotation improves downstream reads and model reliability.¹⁰ A novel cardiac index derived through deep learning adds another option for cardiomegaly detection on routine radiographs.¹¹ Machine learning distinguishes canine intracranial glioma type and grade from Magnetic Resonance Imaging (MRI) texture features that humans do not easily parse.¹² Convolutional networks classify canine round cell tumors on whole-slide images and assist with grading tasks that are time-consuming by eye.¹³

Wearables quantify scratching, head shaking, sleep disruption, and activity in dogs with dermatitis or chronic pain, providing clinics with continuous context for treatment decisions.¹⁴ Owner-facing alert programs tied to device data have shown practical engagement during pruritus treatment.¹⁵ Retrospective work links accelerometer metrics to therapeutic response for canine pruritus, which supports objective monitoring in real care.¹⁶ A collar-based sensor differentiates activity and physiology between osteoarthritic and healthy dogs, paving a path to remote pain tracking in primary care.¹⁷ Emerging comparisons between commercial dog wearables and research-grade accelerometry help teams understand correlation and limits for field deployment.¹⁸ On the text side, NLP has extracted antimicrobial usage patterns from clinical notes at a national scale and fed stewardship dashboards.¹⁹ Text mining pipelines also support disease surveillance from routine veterinary data, demonstrating how narrative fields can power public health.²⁰

Every practice has tried tools that looked promising in a demo, then stalled in the exam room once real constraints appeared. The first barrier is regulatory alignment. In the United States, telemedicine sits within a VCPR except for narrow emergency advice, and clinics must document that relationship clearly.¹ Clinics use AVMA guidance to sketch policies and staff scripts, yet variability by state makes clinics cautious, especially when tools blend education, triage, and treatment reminders.² The United Kingdom’s “under care” guidance clarifies responsibility for remote interactions and prescribing, which helps leaders design safe pathways that fit local norms.³ The prudent stance is to build systems that never step outside those guardrails without a documented examination in the record.

Trust and clarity come next. Clinicians welcome decision support that reduces cognitive load and surfaces risk when the tool shows its evidence, uses familiar language, and stays in assist mode with visible human sign-off.⁵ In imaging, teams respond to models that highlight suspected cardiomegaly on canine thoracic films and pair the output with an overlay that links to a measure such as Vertebral Heart Size (VHS), which converts an abstract score into a familiar clinical cue.⁸ Primary studies that detect left atrial enlargement on radiographs reinforce the value of interpretable outputs that clinicians can verify against their own interpretations. Pathology assistance that flags regions driving a tumor grade or counts mitoses can speed review, yet adoption depends on fast review screens that allow easy override and alignment with reporting standards.¹³

Workflow cost is a third barrier. Tools that require jumping between portals or copying data into the Patient (or Practice) Information Management System (PIMS) add minutes that busy teams cannot spare. Leaders see stronger uptake when imaging assistance appears inside the Picture Archiving and Communication System (PACS), when owner videos land in the visit note with a timestamp, and when any summary carries clickable links back to the source content.⁶ Integration is the difference between a feature used hundreds of times per week and one used during pilots only.

Data quality complicates otherwise strong models. Owner-submitted media files vary widely in angle, lighting, and duration. Clinic radiographs include underexposure, clipped anatomy, and mild rotation, which a quality detector can flag before interpretation.¹⁰ Training technologists with a short quality checklist improves baseline inputs and preserves accuracy for any AI layered on top. Wearables create their own artifacts when collars rotate or leashes attach near the sensor, which calls for simple setup guides and confidence prompts when data appear implausible.

Role clarity and client communication complete the picture. Chat-based helpers can draft histories or after-visit summaries, yet clients need to see that licensed clinicians remain the decision-makers.⁵ Staff need one-sentence ways to explain what the model did and why its suggestion appears, which reduces hesitation and keeps the conversation centered on the pet rather than the tool.⁴ Clinics benefit from model fact sheets that list intended use, inputs, known pitfalls, and contact points for issues, which builds transparency and speeds troubleshooting.⁶

Two brief cases show both promise and friction. A hospital deploys a chest radiograph model that flags cardiomegaly with heatmaps and VHS-aligned cues, which increases reader confidence and shortens reporting. Technologists notice the model struggles on rotated or underexposed films, so a paired quality checker prompts retakes when needed, which restores performance and reduces callbacks.¹⁰ A primary care clinic pilots an intake tool that structures owner narratives, which shortens visits and improves handoffs. When default messages drift toward treatment reminders for patients without an established exam, compliance risks rise and usage dips until settings are corrected to mirror VCPR rules and staff receive new scripts.¹

Leaders can turn caution into confidence by committing to three things: compliance by design, explanations clinicians can scan in seconds, and frictionless workflow. Compliance by design starts with permissions and geography. Configure triage and education tools to default to advice within an existing VCPR, and label pre-VCPR content clearly so clients understand what can and cannot happen remotely.¹ Surface state-aware prompts that show local telehealth rules when a team member schedules a virtual visit, and route cases accordingly. Offer templates that document VCPR establishment and remote follow-ups; they prove alignment at audit time and remove uncertainty for staff. Add consent screens when remote tools collect media or symptoms, then store those consents with the visit to support privacy and informed use.⁴

Explanations must be fast and familiar to fit clinical tempo. Imaging tools should pair predictions with overlays and tie outputs to measures clinicians already use during routine reads, for example, a VHS-aligned cardiomegaly score or a region map showing suspected mass effect.⁷ Primary studies detecting left atrial enlargement on canine radiographs give teams clinical face validity, especially when the interface links directly to the marked region and shows confidence levels in practice.⁸ A pilot reporting accurate LAE detection on routine films underscores the value of tools that summarize with a single cue and a short, legible rationale.⁹ Place quality detection upstream to flag underexposure or rotation before interpretation, keeping downstream reads more reliable. For cardiac workups, a deep-learning cardiac index with clear thresholds adds another usable cue that maps neatly to a General Practitioner’s (GP) decision points.

Workflow fit is the keystone. Embed image review inside the PACS so readers never leave the screen they trust. Integrate owner video ingestion into the PIMS with a simple capture script and an automatic link in the note to preserve provenance and reduce duplication. For referral cases, let MRI texture assistance appear as a side panel that a neuroradiologist can consult during interpretation rather than a separate report, which keeps responsibility clear and speeds use at each step. Pathology support should highlight tiles that drove a grade and provide one-click accept or adjust, preserving operator control and reducing friction.¹³

Measure what matters and share it. Track adoption by role, time saved per visit, retake rates after quality prompts, agreement with specialists on key findings, and client comprehension after AI-assisted education. Publish a simple dashboard for the team and retire features that create extra steps without value. Leaders can start with one imaging assist and one intake helper, then expand once the data show time saved and improved decisions.

Finally, keep transparent model fact sheets in the record. List training data domains, intended use, known blind spots, and contacts for support. Share update notes with staff so changes never arrive as surprises. This cadence builds trust and shortens the path from pilot to everyday tool.⁵

Most disease burden happens away from the clinic, which turns episodic care into a guessing game. A dermatitis flare shows up first as poor sleep and a jump in scratching. A painful morning for an osteoarthritic dog starts as slower movement on stairs before owners mention it. Cardiopulmonary strain can surface as subtle rest changes long before a crisis. Wearables and computer vision can fill the gap, yet programs fail when signals are noisy, thresholds are unclear, and next steps are not wired into staff workflows.

Accelerometers capture activity, rest, and specific behaviors such as scratching or head shaking, which allows objective tracking of pruritus in daily life. Clinics have run owner-facing alert programs tied to device data, which have prompted timely action and gave teams a way to monitor response on days when appointments were not possible.¹⁶ Retrospective work now links accelerometer metrics to treatment response during real care, which supports the use of short monitoring sprints around medication changes.¹⁷ Pain management presents a similar opportunity. A collar-based sensor discriminated activity and physiological variables between dogs with osteoarthritis and healthy controls, which suggests a role for combining activity, rest, and autonomic trends during titration.¹⁸

Noise is inevitable and manageable. Collar rotation, leash attachment near the device, or unusual play patterns can skew readings in ways that look like symptoms. A practical path is to use short setup guides, day-one calibration checks, and low-confidence flags that prompt owners to reseat the collar or send a short validating video. Teams that treat artifacts as solvable input problems get better data and fewer false alarms, which keeps clinicians engaged and clients calmer.¹⁴

Signal-to-action gaps are the next failure point. Many pilots generate pretty trend lines without ladders that define who does what when a threshold trips. A useful dermatitis ladder looks like this: if the three-day scratching index rises 30% over baseline, push skin-care education and offer a nurse slot; if the pattern persists seven days, book the veterinarian for plan revision.¹⁵ An osteoarthritis ladder might combine a four-day drop in night rest with a downward Heart Rate Variability (HRV) trend to trigger a pain check and possible therapy adjustment. Cardiopulmonary ladders can pair owner cough counts from guided videos with rest respiratory rate, then move the patient forward on the schedule when both rise together.¹⁴

Imaging has parallel blind spots. A busy GP might not re-measure VHS on every chest film, and motion or exposure problems can hide subtle changes that matter. Quality detection that flags rotation or underexposure before interpretation improves the reliability of both human reads and any AI layered on top, which protects continuity of care.¹⁰

Owners need feedback that teaches rather than alarms. If an app only shows red dots, anxiety grows. If it explains the trend in one sentence, shows a short example clip of the behavior in question, and pairs the change with a clear next step and a path to a nurse or doctor, adherence improves and escalations arrive earlier. The clinic must own that loop so that data move through education into decisions and better days at home.

A practical home-monitoring program rests on four pillars: validated metrics, short protocols, exception-based workflows, and teaching that sticks.

Start with validated signals and clear device setup. For pruritus, choose devices and algorithms that distinguish scratching and head shaking and have demonstrated sensitivity to treatment response in real life. Create a one-page setup guide with collar fit diagrams, charging tips, and a quick test to confirm detection, which reduces artifacts before they frustrate staff or clients.¹⁵ Add owner-guided cough videos with a capture script when respiratory signs are part of the problem list, proving a consistent visual record that complements sensor data.

Keep protocols tight. Define 14-day sprints that answer focused questions such as whether the dermatitis plan is working or whether an osteoarthritis change improved sleep and mobility.¹⁷ Set a daily completeness goal and schedule a 10-minute nurse call at the midpoint to troubleshoot fit or charging. Close the sprint with a structured note that pulls metrics and a plain-language interpretation into the Electronic Health Record (EHR), which preserves learning for future visits and keeps the narrative grounded in measurable change.

Route by exception instead of streaming raw data. Build ladders into the PIMS so thresholds generate tasks for the right role. For dermatitis, a 30% rise in the three-day scratching index triggers education and a nurse slot, while a seven-day persistence books the veterinarian. For osteoarthritis, combine drops in night rest with HRV trends to prompt a pain review and possible therapy adjustment.¹⁸ For cardiopulmonary cases, link owner cough counts from guided videos to rest respiratory rate; when both rise, move the patient forward on the schedule and prefill imaging requests if indicated.¹⁶

Teach owners with rhythm and clarity. When a threshold trips, send a single sentence that explains the pattern, a short clip that labels what counts as scratching or coughing, and a clear next step with a link to schedule.¹⁴ Provide a clinic-branded overview of what the device measures, what it does not, and how data privacy works, which builds confidence and reduces churn. Make device pickup and return easy with a barcoded loaner closet and QR-coded instructions, which saves staff time and keeps hardware circulating.

Measure results and celebrate wins. Track sprint completion rates, the share of exceptions that led to medication changes, time to itch control after therapy changes, and unplanned visits avoided.¹⁴ Review outcomes monthly and expand signals only when you can show an action pathway and demonstrated outcome improvement. Publish short success stories inside the team so protocols become habits rather than one-off projects.

Veterinary medicine runs on narratives. The details that break a case often sit in free text: intermittent vomiting after boarding, a cough that worsens with excitement, a household change that preceded a flare. These notes are gold for continuity and stewardship, yet they are hard to search and summarize during busy days. NLP can help when it respects context, preserves nuance, and feeds decisions clinicians actually need to make.¹⁹

Primary care differentials are broad and noisy. Early Cushing’s disease looks like many problems during the first suspicion visit. A careful study used machine learning on UK primary care records to predict a veterinarian’s recorded diagnosis of Cushing’s at the point of first suspicion, which achieved strong discrimination and offered a template for syndrome-level triage that reflects real documentation.¹⁹ Stewardship depends on seeing prescribing patterns across large pools of notes. Work with national clinical datasets shows that NLP can extract antimicrobial usage with enough fidelity to support dashboards that track indication, dose, and duration at scale.

Pathology and imaging notes pose related hurdles. Whole-slide images now support models that classify canine round cell tumors and assist with grading tasks such as mitotic counts, which can draft structured language that aligns with standards and reduces time spent on repetitive tasks.¹³ Radiology reports benefit when findings map to structured terms and link to prior comparisons, which allows groups to answer simple but important questions such as how often a “suspicious for” read leads to re-imaging or how long it takes to move from suspected cardiomegaly on radiographs to an echocardiogram referral.⁷

Governance ties it together. Clinics need clear policies for model selection, versioning, error reporting, and delineation of use. A global framework emphasizes safety, transparency, and accountability, which translates into simple local practices such as model registries, shadow mode trials before activation, and weekly review of overrides in cases where a model and a clinician disagreed.⁴ The goal is to convert messy text and pixels into timely, auditable decisions while preserving clinician judgment and patient context.

The challenge is the conversion of narrative and image data into decisions that reduce callbacks, support stewardship, and produce clear follow-ups without extra clicks. That requires respect for clinical roles, opt-in designs that keep humans in control, and a commitment to measuring whether NLP and imaging support improve care pathways in ways that teams can feel during the workday.

Start with the questions clinicians ask most and build tools that answer them in one screen. For primary care, that usually means triage, likely differentials, and next best action. Create NLP pipelines that transform intake narratives into a structured snapshot with duration, triggers, red flags, prior therapies, and environmental changes, then place that snapshot at the top of the note. Keep every line clickable back to the source sentence so the veterinarian can verify quickly and maintain confidence in the summary.²⁰

Build stewardship from everyday documentation. Extract antimicrobial name, dose, duration, indication, and culture references from notes and semi-structured fields into a clinic dashboard. Begin with a few high-value syndromes and review accuracy with a rotating panel of clinicians. Share simple monthly views that show the percentage of cases with culture when indicated, broad-spectrum usage trends, and top indications by service. Use the dashboard in huddles to celebrate progress and identify training needs.

Embed image assistance where work happens. For thoracic radiographs, pair a quality detector with a cardiomegaly assist so technologists correct exposure or rotation before reads and clinicians see an overlay plus a VHS-aligned score with confidence.⁷ Train staff with short modules on common artifacts and positioning, then track retake rates and agreement with specialist reads. Referral centers can pilot MRI texture analysis for suspected gliomas with neuroradiology oversight, logging cases where the model sharpened differentials or shortened time to decision.

Support pathology with review screens that make expert judgment faster. Deploy models that assist with grading and mitotic counts on canine tumors, then auto-populate a structured summary once the pathologist approves or adjusts. Standardize wording and codes across sites so downstream analytics transfer and so clinicians can search across cases for patterns that matter to treatment.

Governance and measurement keep everything safe and useful. Create a model registry that lists intended uses, validation notes, and known failure modes for each tool. Run new features in shadow mode for two weeks before turning them on for care decisions, which lets teams compare outputs with human decisions without risk. Log overrides and disagreements, then review a sample weekly to spot drift early. Publish clinic metrics on time saved, callback reduction, stewardship adherence, and imaging retake rates. Retire features that do not move outcomes and expand those that do.¹⁹

When notes and pixels become decisions that teams trust, AI stops feeling like a science project and starts feeling like good medicine. Clinicians see faster reads, cleaner handoffs, and fewer returns for confusion, while owners experience clearer guidance and earlier interventions that improve daily life for their animals.

Evidence strength varies by task, so match the tool to the job. Imaging-led use cases are furthest along: radiology assistance for canine cardiomegaly, quality detection, and several pathology classifications show strong technical performance, even as many clinical endpoints still need confirmation as systems move from pilots into day-to-day care. Wearables add valuable signal on pruritus behaviors, sleep disruption, and pain-related activity shifts, but real-world quirks, attachment artifacts, and leash handling, can skew readings, and devices aren’t interchangeable without local calibration. 

NLP can reliably surface stewardship and triage cues when documentation habits support it, yet accuracy remains context-dependent and should be validated on-site before activation. Generative tools are useful for supervised summaries and education, provided governance prevents scope drift and that it manages hallucination risk. Because telemedicine and prescribing rules vary and change, compliance-first, auditable designs are the safest default. Clinics that anchor on high-evidence use cases, instrument outcomes, and iterate quickly tend to see steadier gains with fewer surprises.

AI becomes real in pet care when it helps teams do what they already value. This piece mapped three stubborn obstacles and three practical opportunities. First, compliance, trust, and workflow friction stall adoption. Build VCPR-safe tools that explain themselves in a sentence and live inside the record.¹ Second, blind spots between visits hide flares and slow response. Use validated sensors and short playbooks so clinicians see meaningful change at the right time and route action to the right role. Third, unstructured notes and complex differentials slow decisions. Turn narratives and images into auditable, searchable insights for triage, stewardship, imaging, and pathology.

Leaders can move now. Pick one imaging assist and one home-monitoring sprint. Wrap them in clear guardrails, one-page fact sheets, and good measurement. Align with AVMA guidance and with global principles for ethics and transparency in health AI. Celebrate time saved and earlier interventions in monthly huddles. Share lessons across sites. Expand when outcomes improve and staff report that the tools removed steps rather than adding them.

The Decision Lab partners with veterinary groups, referral centers, and animal health companies to translate behavioral science and AI into patient outcomes. We design compliance-first workflows, build explainable interfaces, and wire sensors and models to playbooks that teams enjoy using. If your mandate is to make care faster, clearer, and more continuous, we can help you turn scattered data into reliable decisions. Reach out, and let’s build pet care that feels modern, humane, and effortless for teams and families alike.

How Personalizing Nudges Could Change Health Care

A fast, practical look at tailoring reminders and follow-ups so the right message reaches the right patient at the right moment—useful patterns for pet-owner alerts, triage scripts, and escalation rules.

The New Personalized AI Nutritionist

What it takes to turn AI insights into daily choices people actually follow. Helpful for designing explainable recommendations, consent flows, and feedback loops that mirror veterinary workflows. 

1. American Veterinary Medical Association. (2021). Veterinary telehealth: The basics. https://www.avma.org/resources-tools/animal-health-and-welfare/telehealth-telemedicine-veterinary-practice/veterinary-telehealth-basics
2. American Veterinary Medical Association. (2021, January 21). Telehealth guidelines give veterinarians a place to start. Journal of the American Veterinary Medical Association News. https://www.avma.org/javma-news/2021-02-01/telehealth-guidelines-give-veterinarians-place-start
3. Royal College of Veterinary Surgeons. (2023, January 20). RCVS Council approves new guidance on “under care” and 24/7 cover. https://www.rcvs.org.uk/news-and-views/news/rcvs-council-approves-new-guidance-on-under-care-and-247-cover/
4. World Health Organization. (2021). Ethics and governance of artificial intelligence for health. https://www.who.int/publications/i/item/9789240029200
5. Chu, C. P., Chen, C. H., & Chen, W. C. (2024). ChatGPT in veterinary medicine: A practical guidance of applications and precautions. Frontiers in Veterinary Science, 11, 1395934. https://doi.org/10.3389/fvets.2024.1395934
6. Abu-Seida, A. M. (2024). Veterinary telemedicine: A new era for animal welfare. Veterinary World, 17(3), 457–463. https://pmc.ncbi.nlm.nih.gov/articles/PMC11128645/
7. Burti, S., Banzato, T., Coghlan, S., Wodzinski, M., Bendazzoli, M., & Zotti, A. (2024). Artificial intelligence in veterinary diagnostic imaging: Perspectives and limitations. Research in Veterinary Science, 175, 105317. https://doi.org/10.1016/j.rvsc.2024.105317
8. Burti, S., Longhin Osti, V., Zotti, A., & Banzato, T. (2020). Use of deep learning to detect cardiomegaly on thoracic radiographs in dogs. The Veterinary Journal, 262, 105505. https://doi.org/10.1016/j.tvjl.2020.105505
9. Li, S., Tsai, S., Hsu, W. L., Wang, P. H., & Chou, P. Y. (2020). Pilot study: Application of artificial intelligence for detecting left atrial enlargement on canine thoracic radiographs. Veterinary Radiology & Ultrasound, 61(6), 611–618. https://doi.org/10.1111/vru.12901
10. Banzato, T., Wodzinski, M., Burti, S., Vettore, E., Müller, H., & Zotti, A. (2023). An AI-based algorithm for the automatic evaluation of image quality in canine thoracic radiographs. Scientific Reports, 13, 17024. https://doi.org/10.1038/s41598-023-44089-4
11. Jeong, Y., & Sung, J. (2022). An automated deep learning method and novel cardiac index to detect canine cardiomegaly from simple radiography. Scientific Reports, 12, 14494. https://doi.org/10.1038/s41598-022-18822-4
12. Barge, P., Oevermann, A., Maiolini, A., & Durand, A. (2023). Machine learning predicts histologic type and grade of canine gliomas based on MRI texture analysis. Veterinary Radiology & Ultrasound: The Official Journal of the American College of Veterinary Radiology and the International Veterinary Radiology Association, 64(4), 724–732. https://doi.org/10.1111/vru.13242
13. Salvi, M., Molinari, F., Iussich, S., Muscatello, L. V., Pazzini, L., Benali, S., Banco, B., Abramo, F., De Maria, R., & Aresu, L. (2021). Histopathological classification of canine cutaneous round cell tumors using deep learning: A multi-center study. Frontiers in Veterinary Science, 8, 640944. https://doi.org/10.3389/fvets.2021.640944
14. Wernimont, S. M., D’Aniello, B., Payne, K., & Bartges, J. W. (2018). Use of accelerometer activity monitors to detect changes in pruritic behaviors in dogs. Sensors, 18(1), 249. https://doi.org/10.3390/s18010249
15. Carson, A., Vukovic, K., & Marsella, R. (2023). Response of pet owners to Whistle FIT activity monitor alerts of pruritic behaviors in dogs. Frontiers in Veterinary Science, 10, 1123266. https://doi.org/10.3389/fvets.2023.1123266
16. O’Rourke, A., Johnson, K., & Sethi, S. (2025). Retrospective observational study shows accelerometers can objectively monitor response to treatment for canine pruritus. American Journal of Veterinary Research, 86(3), 235–243. https://doi.org/10.2460/ajvr.24.09.0269
17. Rowlison de Ortiz, A., Belda, B., Hash, J., Enomoto, M., Robertson, I., & Lascelles, B. D. X. (2022). Initial exploration of the discriminatory ability of a collar-based sensor to detect differences in activity and physiological variables between healthy and osteoarthritic dogs. Frontiers in Pain Research, 3, 949877. https://doi.org/10.3389/fpain.2022.949877
18. Hilborn, E. C., Rudinsky, A. J., & Kieves, N. R. (2024). Commercially available wearable health monitors in dogs only had a very strong correlation during longer durations of time: a pilot study. American Journal of Veterinary Research, 85(10), ajvr.24.06.0162. https://doi.org/10.2460/ajvr.24.06.0162
19. Hur, B., Hardefeldt, L. Y., Verspoor, K., Baldwin, T., & Gilkerson, J. R. (2019). Using natural language processing and VetCompass to understand antimicrobial usage patterns in Australia. Australian Veterinary Journal, 97(8), 298–300. https://doi.org/10.1111/avj.12836
20. Davies, H., James, S., & Gaskell, R. (2024). Text mining for disease surveillance in veterinary clinical data: A mini-series part two. Frontiers in Veterinary Science, 11, 1352726. https://doi.org/10.3389/fvets.2024.1352726