This study evaluated the performance of the KU-F40 automated analyzer in a routine clinical setting, revealing moderate to substantial agreement (Cohen’s kappa, 0.608) between the KU-F40 and FECT results. Notably, the KU-F40 system exhibited high specificity (99.4%) and NPV (99.8%), confirming its potential to effectively rule out negative samples from low-prevalence populations. However, the system exhibited a relatively low sensitivity of 57.1%, and its PPV was only 29.9% (adjusted for our institutional prevalence of 0.45%). These findings suggest that while the KU-F40 system may be useful as an adjunctive workflow tool in low-prevalence laboratories, it should not replace FECT when high diagnostic sensitivity or precise species-level identification is required.

FECT achieved an overall infection detection rate of 2.08% (21/1,011), slightly higher than the rate of 1.78% (18/1,011) achieved by the KU-F40 system. In contrast, Huang et al. [10] found that an automated system achieved a significantly higher detection rate (8.74%) than manual microscopy (2.81%). This discrepancy likely stems from the fact that Huang et al. [10] used routine direct saline smears, a method that does not involve concentration of the samples and that uses significantly smaller sample volumes than FECT. Here, however, we utilized FECT as a more rigorous reference standard. Consequently, our results are more consistent with those of Boonyong et al. [12], who reported that FECT identified more positive samples than an automated analyzer (468 vs. 440 positive cases, respectively, for FECT and the automated system, among 800 preserved samples) [12]. This disparity is primarily because the manual concentration procedure of FECT uses a larger stool volume than the automated analyzer and because FECT effectively removes fat and debris using ethyl acetate. In summary, the KU-F40 system provides as a practical solution to bridge the gap between traditional methods. While routine direct smear analysis is convenient, it carries a high risk of false negatives. Although FECT provides superior sensitivity, its high labor intensity and risk of exposure to hazardous reagents present significant drawbacks.

The operational efficiency of the KU-F40 system was evaluated alongside that of FECT. The KU-F40 system reduced hands-on labor time and chemical exposure, although its total instrument turnaround time varied according to scan settings and was not necessarily shorter than FECT. In clinical laboratories, the KU-F40 system may complement traditional methods. Here, while FECT identified 21 positive cases, a total of 27 samples were positive by at least one of the two methods. Our findings reveal that the KU-F40 system can identify low-density positive cases, such as those with low-density C. sinensis infection, that may be overlooked during manual FECT. Considering that C. sinensis is the most prevalent and medically important parasite in South Korea, the ability of this system to detect low-density infections is of paramount clinical significance.

The system may have advantages for detecting certain helminth eggs, particularly when images are captured at 20× magnification, but species-specific sensitivity could not be estimated reliably because of the small number of positive samples. This potential benefit likely stems from the multilayer scanning protocol which mimics manual fine focusing [13]. While 20× magnification is sufficient to recognize the relatively large and distinct morphological features of helminth eggs, it limits the fine differentiation required for identifying and distinguishing between small and similar protozoa [10,13]. In settings with a high prevalence of protozoa, adjusting the parameters to acquire more images at 40× magnification may improve detection. Further studies across diverse settings are required to determine the optimal imaging parameters. Our finding of exceptionally low species-level concordance (only six cases) suggests that the AI training database requires further expansion and refinement. Furthermore, owing to the limited resolution and image quality, the objects in the images exhibit ambiguous or atypical morphologies. Such suboptimal imaging leads to an interpretive gap between automated categorization and expert review, potentially challenging even experts in performing accurate diagnosis. Therefore, while the KU-F40 system requires expert validation to reach its full potential, expert validation cannot fully compensate for the diagnostic uncertainty caused by suboptimal image quality.

Our study has several limitations. First, it was conducted at a single tertiary center in an urban area, where the institutional prevalence was 0.45%, and where most of the patients undergoing parasite testing were asymptomatic individuals undergoing routine health checkups. This geographic bias limited the number of positive specimens, and the study population may not exhibit the higher prevalence or epidemiological profiles of rural or riverside populations in South Korea. Second, the use of a localized study sample limited the diversity of parasite species in the samples. Although various species were identified, most of the positive cases exhibited one of a few prevalent species, such as M. yokogawai. Species endemic to specific regions, such as G. seoi, which is commonly found in river basins, were underrepresented. Therefore, the KU-F40 system’s performance for a broader spectrum of clinically significant helminths or protozoa could not be extensively evaluated. Future multicenter studies across diverse geographical regions are essential to provide the necessary datasets for further AI algorithm training and to validate the KU-F40 system’s performance in various epidemiological contexts. Third, FECT was used as the routine reference standard, although it is operator-dependent and may not represent an error-free gold standard. Unlike the KU-F40 system, in which high-resolution digital images are archived for retrospective review, manual FECT slides cannot be reverified in cases of discrepancy. This inability to confirm the discordant results from manual examination may have led to a conservative estimation of the diagnostic performance of the KU-F40 system. Further, the presence of ambiguous or atypical morphology in images captured by the KU-F40 system can challenge expert review and confirmation, particularly when only a single image is available. Here, for instance, an organism identified as Ascaris lumbricoides by the KU-F40 system was found to be as T. trichiura based on FECT. This case highlights the need for confirmatory testing via manual methods such as FECT or via molecular methods when image review is inconclusive.