This is the second study demonstrating that the VETSCAN IMAGYST system integrated with a deep learning object detection algorithm can successfully recognize and identify diagnostic forms of gastrointestinal parasites in dogs and cats on fecal flotation slides scanned by an automated microscope. Whereas our previous study evaluated the VETSCAN IMAGYST system for the detection of eggs of Ancylostoma, Toxocara, Trichuris, and Taeniidae in 84 canine and 16 feline fecal samples [40], this study assessed the ability of the system to detect eggs of the feline nematodes Ancylostoma and Toxocara cati, and oocysts of the protozoan parasite Cystoisospora and cysts of the protozoan parasite Giardia, in 104 canine and 96 feline fecal samples, making it a more comprehensive analysis of this novel system.
Although both domestic dogs (Canis lupus familiaris) and cats (Felis catus) belong to the order Carnivora, dogs are classified into the superfamily Canoidea and cats are classified into the superfamily Feloidea [46, 47]. Diets of canids can vary from herbivorous to omnivorous; however, all felids require a strictly carnivorous diet [47]. A high-protein diet of animal origin is essential for domestic cats to obtain some of their nutritional requirements, such as taurine, as well as arachidonic acid and vitamin A [47]. Due to their diet, fecal samples of cats commonly contain a large amount of fat and are soft, sticky and clay-like in consistency, which often makes it more difficult, or sometimes impossible, to read fecal slides since more debris floats with fats, especially when a viscous sugar solution is used in a centrifugal flotation technique. Modification of fecal flotation procedures, involving an initial water wash where the supernatant is discarded after the initial spin and the sediment resuspended with a flotation solution to remove excess fat, mucus, and debris (double centrifugal fecal flotation technique), can be used [10]. In the present study, the VETSCAN IMAGYST centrifugal flotation method recovered parasite elements from feline feces, and the VETSCAN IMAGYST scanner and algorithm successfully captured and identified targeted parasites (Fig. 5).
To our knowledge, this is the first report demonstrating the ability of the VETSCAN IMAGYST system to recover and accurately detect protozoan parasites, i.e. Cystoisospora oocysts and Giardia cysts. Although coccidiosis is generally considered a self-limiting infection in mature dogs and cats due to their rapid development of immunity [18], Cystoisospora is an ubiquitous and important pathogen in puppies and kittens, with infection often resulting in diarrhea, abdominal pain, anorexia, bloody diarrhea, anemia, and even mortality in severe cases [10, 17, 48]. Since Cystoisospora undergo fast replication in the pathogenic intestinal stage and a high number of oocysts are excreted in host feces, causing environmental contamination, it is considered critical and thus highly recommended to conduct a fecal examination with centrifugation for puppies and kittens at least four times during the first year of life for treatment of Cystoisospora at an early stage of infection [48, 49]. Different species of Cystoisospora are commonly diagnosed in dogs and cats: Cystoisospora canis and Cystoisospora ohioensis in dogs, and Cystoisospora felis and Cystoisospora rivolta in cats. Oocysts of C. canis and C. felis are slightly bigger, at approximately 38-51 × 27-39 µm in size, than those of C. ohioensis and C. rivolta, which are approximately 17-27 x 15-24 µm in size [10]. Due to the smaller size of coccidian oocysts compared to helminth eggs, Cystoisospora can be easily overlooked, especially when the number of oocysts on a fecal slide is low and an inaccurate microscopic focus is used for examination. The VETSCAN IMAGYST system correctly identified oocysts of all four Cystoisospora species in canine and feline fecal samples and successfully reported them as Cystoisospora (coccidia) (Fig. 5).
The diagnostic sensitivity and specificity of the VETSCAN IMAGYST scanner and algorithm for the Giardia samples compared with the results reported by the experts were 75.8 and 97.0%, respectively (Table 2). As previously discussed, a common challenge for many object detection algorithm models is to precisely localize and distinguish small objects such as Giardia cysts [40]. The nature of a deep learning algorithm, however, means that its performance continues to improve with further training. It is important to note that the diagnostic sensitivity was dramatically increased to 95.2% by removing the 12 Giardia samples that had ≤ 10 CPG from the analysis; detecting such a low number of cysts is extremely demanding when slides are examined by visual microscopy. Additionally, the examination and counts of CPG on these Giardia slides were carefully performed by a well-trained diagnostic parasitologist with no time limit, which most likely resulted in a much higher diagnostic performance compared to that usually achieved by technicians in daily veterinary practice.
The detection of Giardia cysts and triphozoites by microscopic examination is generally considered the most sensitive technique for the diagnosis of giardiasis, and therefore, a great deal of training and experience is required for the confident diagnosis of this disease [10]. It is challenging to identify Giardia infection by fecal examination because, in addition to their small size and transparency, Giardia cysts and trophozoites are intermittently shed in feces, and multiple fecal examinations may be necessary to rule out infection. Fresh fecal samples, obtained preferably within 30 min of defecation, are often required to detect motile trophozoites; however, Giardia cysts and trophozoites are fairly fragile and their shape easily distorted in flotation solution [10]. A 33% zinc sulfate solution (specific gravity, 1.18) is preferred and recommended for the detection of Giardia cysts, as other flotation solutions can rapidly cause osmotic damage to them, which increases the difficulty of perceiving them on fecal slides [10, 34, 50, 51]. During the present study, the VETSCAN IMAGYST system effectively recognized and identified both intact and collapsed Giardia cysts (Fig. 5). Testing for Giardia is recommended not only in symptomatic dogs and cats, but also in dogs and cats newly introduced to homes which have other pets that are free of infection, as many Giardia infections can be asymptomatic [10, 32, 48]. Since there is no perfect flotation solution for the recovery of all the different types of parasites [10], it is important to consider the advantages and disadvantages of each individual solution when selecting one for general use. Some experts recommend performing two centrifugal flotation tests by using both Sheather’s sugar and 33% zinc sulfate solutions to achieve a broader range of gastrointestinal parasite detections. In cases where Giardia is suspected, analysis using the sugar flotation solution should also be performed on fecal samples to check for other parasites.
The detection of Giardia is also possible with Giardia-specific coproantigen detection assays [35, 48]. However, when not used in conjunction with a traditional microscopic technique, antigen testing may provide a false positive result in an animal that is no longer infected with Giardia due to persistent antigen excretion for several weeks or even months after parasite elimination [52, 53]. Given the shortcomings of current in-house diagnostic methods for Giardia, utilizing a deep learning algorithm platform, such as the VETSCAN IMAGYST system, could provide clinicians with an excellent additional or alternative diagnostic tool to help identify Giardia cases that would otherwise be missed.
Evaluation of the performance of the VETSCAN IMAGYST centrifugal flotation sample preparation method was limited in this study due to the modest numbers of true positives for the four targeted parasites and the inherent subsampling variability in non-homogenous fecal samples, which has been well documented in previous publications [54]. Kochanowski et al. [54] observed a wide range of coefficients of variation, between 31 and 254%, for Toxocara and Trichuris samples with a low number of egg counts, i.e. ≤ 50 EPG. Despite these limitations, the performance of the VETSCAN IMAGYST centrifugal flotation method in the present study was comparable to a conventional centrifugal flotation method, with diagnostic sensitivity and specificity of the comparisons ranging from 65.7 to 100% and 97.6–100%, respectively, across the four targeted parasites (Table 3). Additionally, one potential modification considered for the VETSCAN IMAGYST centrifugal flotation method to increase its diagnostic sensitivity is to lengthen the duration of centrifugation. Previous studies reported that egg recoveries with centrifugation at 264 × g were significantly improved when the duration of centrifugation was increased from 1 and 3 min to 4 or 5 min at the same speed, although no change was observed in egg recovery when the time of centrifugation was extended to 10 or 20 min [37, 55].
As shown in Table 3, the diagnostic sensitivity and specificity of the VETSCAN IMAGYST centrifugal technique slightly surpassed those of the OVASSAY passive flotation method. Despite the fact that centrifugation significantly increases the sensitivity of fecal examinations, passive flotation continues to be the most commonly used technique in veterinary private practice due to its convenience [10, 19, 36, 51, 56,57,58]. Given that the VETSCAN IMAGYST system reliably recovers and detects parasite elements in fecal samples, does not largely depend on the experience level of examiners, and has previously been shown to provide results in around 10 min with the VETSCAN IMAGYST centrifugal flotation method [40], it has the potential to replace the conventional passive flotation method used in veterinary practice.
The most distinctive and unique feature of the VETSCAN IMAGYST system is its deep learning object detection algorithm. To the best of our knowledge, the VETSCAN IMAGYST system is the only automated diagnostic system that is integrated with a deep learning object detection algorithm and applied to veterinary medicine. Compared to shallow learning systems, which do not have any structural information on the function to be learned, deep learning algorithms exploit the advantage of locality at each level of the layered hierarchy, enabling the system to ignore the aspects that make computer vision brittle [59, 60]. The layered hierarchy also facilitates the system to continuously adapt to new data and apply these to new output classes with fewer examples [60, 61]. The deep learning characteristic, along with the YOLOv3 object detection model [43], which incorporates localization and classification features, results in a decrease of background errors and high agreement between the VETSCAN IMAGYST system and expert examinations. Another benefit of this system is its ability to store images and reports on a secure, cloud-based server system, allowing easy sharing of information by parasitologists as well as members of the veterinary and academic communities for patient care, research, and teaching.
The present study did not evaluate the usability of the VETSCAN IMAGYST system; however, our previous analysis showed that the VETSCAN IMAGYST system with the VETSCAN IMAGYST centrifugal flotation method could produce results in around 10 min, which is comparable to conventional fecal flotation tests. This time frame included the time to prepare the sample, i.e. approximately 3.5 min including the 2-min centrifugal incubation time, and the time to scan the images, i.e. approximately 6–7 min [40]. Data from the present study add to the body of evidence demonstrating the performance of the VETSCAN IMAGYST system in detecting intestinal parasite elements recovered from fecal samples. In addition to identifying the protozoan parasites Cystoisospora and Giardia, results from our present and previous studies show the system’s reliable performance in detecting four different genera/group of gastrointestinal parasites (Ancylostoma, Toxocara, Trichuris, Taeniidae) in dogs and cats [40]. With further training, the VETSCAN IMAGYST system will gain the ability to identify other parasites. The quantitative capability of the VESTSCAN IMAGYST system is currently under development. It is predicted that the algorithm will be able to perform a fecal egg counting test in the future.