AI Fecal Test Matches Expert Microscopy for Detecting Giardia in Diarrheic Dogs

Giardia duodenalis is one of the most familiar antagonists in companion animal medicine, a single-celled intestinal parasite that colonizes the small intestine of dogs, triggers watery or mucoid diarrhea, and sheds tough, environmentally resistant cysts that can survive for weeks or months in soil and kennels. Diagnosing the infection has long depended on a handful of laboratory techniques, each with its own trade-offs between sensitivity, cost, equipment requirements, and the expertise of the person peering into the microscope. A new study from researchers at Colorado State University, published in the journal Parasites & Vectors, has now put one of the newest diagnostic tools on the block, an artificial intelligence-enabled point-of-care imaging system, head to head with the traditional gold standards of fecal parasitology, and the results suggest that machine-assisted diagnosis is closing the gap with human experts.

The research team, led by Michelle Larsen and including McKenna Willis, Valeria Scorza, Sangeeta Rao, and Michael Lappin from the Department of Clinical Sciences and the Center for Companion Animal Studies at Colorado State University, set out with a clear primary objective: to compare Giardia cyst detection results among several traditional fecal flotation techniques and an AI-enabled cyst assay, using fecal samples from shelter dogs with diarrhea. The traditional methods included zinc sulfate centrifugation with expert read, sugar centrifugation with expert read, and direct fluorescent antibody testing, known as DFA, also with expert read. The AI platform under evaluation was the Vetscan Imagyst system manufactured by Zoetis, which is designed for in-clinic use and can detect Giardia cysts in approximately ten minutes using machine-learning algorithms trained to recognize the distinctive morphology of the cysts in digitized images of fecal preparations.

The choice of comparator methods reflects the practical realities of veterinary diagnostics. Centrifugal fecal flotation remains the workhorse of most small animal clinics because it is inexpensive and requires only basic equipment, but its performance varies with the flotation solution used. Zinc sulfate centrifugation is widely regarded as one of the better flotation methods for recovering Giardia cysts, while sugar centrifugation is commonly used for the broader detection of gastrointestinal parasites. DFA, by contrast, uses fluorescently labeled antibodies that bind specifically to Giardia cysts, making it one of the most sensitive detection methods available, but it demands specialized fluorescence microscopy and trained personnel, which places it beyond the reach of most general practices. By anchoring the comparison to DFA and to expert-read flotation, the investigators created a benchmark that any point-of-care technology would have to approach closely to be considered clinically credible.

The study’s headline numbers are instructive. When DFA served as the reference standard, the AI-enabled Giardia cyst assay achieved a sensitivity of 71.4 percent and a specificity of 94.8 percent. Zinc sulfate centrifugation with expert read, measured against the same DFA benchmark, posted an identical sensitivity of 71.4 percent but a perfect specificity of 100 percent. When the two non-DFA methods were compared directly with each other, zinc sulfate centrifugation with expert read versus the AI-enabled cyst assay, the AI system delivered 80.0 percent sensitivity and 89.4 percent specificity. In other words, the machine-driven assay performed in roughly the same diagnostic territory as the expert-performed flotation technique that most clinics already rely on, and neither matched the analytic sensitivity of DFA, which detects cysts that both other methods can miss.

One of the most technically interesting parts of the study involved the use of polymerase chain reaction, or PCR, to adjudicate the discordant results. There were three samples that tested positive by DFA but negative by the AI-enabled Giardia cyst assay, and PCR confirmed that all three of these were genuinely false negatives on the AI side. Conversely, there were three samples that were negative by DFA but positive by the AI-enabled assay, and PCR confirmed that two of these were actually false negatives by DFA, meaning the AI system had correctly flagged infections that the antibody-based reference method had missed. This kind of molecular tie-breaking is a valuable feature of modern diagnostic studies, because it reveals that no single method is infallible and that apparent errors on one platform may in fact reflect limitations of the comparator rather than of the technology being tested.

The investigators also evaluated an AI-enabled fecal ova and oocyst assay, a broader screening mode intended to detect a range of gastrointestinal parasites rather than Giardia specifically, as a Giardia detection tool. This assay produced the lowest sensitivity for Giardia cysts among the methods tested, and the team concluded that it should not be used for Giardia detection, reserving that role for the dedicated Giardia cyst assay mode. The distinction matters operationally: the same imaging platform can run different algorithmic modes optimized for different targets, and the study demonstrates that performance characteristics do not necessarily transfer between modes. A general parasite screen that performs well for hookworm eggs or roundworm ova may not be tuned finely enough to catch Giardia cysts, which are smaller and morphologically subtler than many helminth eggs.

A secondary objective of the study used the same sample set to verify earlier published findings on other common gastrointestinal parasites, comparing AI-enabled parasite detection with sugar centrifugation read by experts. The results for detecting these other parasites, which include hookworms of the genus Ancylostoma, roundworms of the genus Toxocara, the whipworm Trichuris vulpis, and coccidian parasites of the genus Cystoisospora, were similar to those reported in prior studies. This replication is significant because it suggests that the earlier positive evaluations of AI-assisted fecal screening were not artifacts of a particular sample population or laboratory environment, but reflect a reproducible level of performance across different cohorts of diarrheic dogs.

From a veterinary practice standpoint, the implications are straightforward. Most small animal clinics do not employ a parasitology expert, and few have the fluorescence equipment needed to run DFA in house. Samples that require those capabilities are typically shipped to reference laboratories, adding days of delay during which an infected dog may continue to shed cysts into the environment and potentially infect other animals or, in the case of zoonotic concerns around Giardia, pose questions for the household. An AI-enabled system that returns a Giardia result in roughly ten minutes, with agreement comparable to expert centrifugal flotation, offers a way to move accurate parasitological diagnosis into the exam room, enabling same-visit treatment decisions and faster infection control in shelters, boarding facilities, and multi-dog households.

The authors conclude that the overall results from the three principal methods were similar, and they suggest that the AI-enabled Giardia cyst assay could be considered for use in small animal clinics for Giardia and other parasite detection, particularly in settings where a parasite expert or DFA testing is not routinely available. That framing is appropriately measured. The study does not claim that the AI system outperforms expert microscopy or DFA; rather, it demonstrates that the technology achieves clinically acceptable agreement with established methods while removing the dependence on scarce human expertise. For clinics that already have access to skilled microscopists and reference laboratory support, the traditional methods remain fully viable, and DFA retains its place as the most sensitive option when the clinical stakes demand maximal detection.

The study also illustrates the evolving relationship between commercial diagnostic developers and academic veterinary research. Zoetis provided the AI-enabled device and its supplies for the study, and the corresponding author, Michelle Larsen, is currently a full-time employee of Zoetis while completing a master’s program at Colorado State University; the company supplied the instruments but, according to the published declarations, did not influence study design, data analysis, or other portions of the research. Transparency about such arrangements is essential as AI diagnostics proliferate, and the open-access publication allows independent readers to scrutinize the sensitivity and specificity figures for themselves. As machine-learning diagnostic platforms continue to spread through veterinary medicine, studies of this kind, which benchmark them rigorously against expert-read traditional methods and use PCR to resolve discordant results, will be the standard against which their clinical adoption is judged.

Subject of Research: Comparison of AI-enabled and traditional diagnostic methods for detecting Giardia and other enteric parasites in diarrheic dogs

Article Title: Comparison of traditional and artificial intelligence methods for detection of enteric pathogens in feces of dogs with diarrhea

Article References: Larsen, M., Willis, M., Scorza, V., Rao, S., & Lappin, M. (2026). Comparison of traditional and artificial intelligence methods for detection of enteric pathogens in feces of dogs with diarrhea. Parasites & Vectors. https://doi.org/10.1186/s13071-026-07685-7

Image Credits: AI Generated

DOI: 10.1186/s13071-026-07685-7

Keywords: Giardia duodenalis, artificial intelligence, veterinary diagnostics, fecal flotation, direct fluorescent antibody, PCR, dogs, point-of-care testing, intestinal parasites, Parasites & Vectors, Colorado State University, diagnostic sensitivity