Artificial intelligence-based digital pathology for the detection and quantification of soil-transmitted helminths eggs

Conventional microscopy of Kato-Katz (KK1.0) thick smears, the primary method for diagnosing soil-transmitted helminth (STH) infections, has limited sensitivity and is error-prone. Artificial intelligence-based digital pathology (AI-DP) may overcome the constraints of traditional microscopy-based diagnostics. This study in Ucayali, a remote Amazonian region of Peru, compares the performance of AI-DP-based Kato-Katz (KK2.0) method to KK1.0 at diagnosing STH infections in school-aged children (SAC). In this prospective, non-interventional study, 510 stool samples from SAC (aged 5-14 years) were analyzed using KK1.0, KK2.0, and tube spontaneous sedimentation technique (TSET). KK1.0 and KK2.0 slides were evaluated at 30-minute and 24-hour timepoints for detection of Ascaris lumbricoides, Trichuris trichiura, and hookworms (at 30-minute only). Diagnostic performance was assessed by measuring STH eggs per gram of stool (EPG), sensitivity of methods, and agreement between the methods. KK2.0 detected more A. lumbricoides positive samples than KK1.0, with detection rates for T. trichiura and hookworms being comparable. At 30-minutes, 37.6%, 23.0%, and 2.6% of the samples tested positive based on KK1.0 for A. lumbricoides, T. trichiura, and hookworms, while this was 49.8%, 24.4%, and 1.9% for KK2.0. At 24-hours, 37.1% and 27.1% of the samples tested positive based on KK1.0 for A. lumbricoides and T. trichiura, while this was 45.8% and 24.1% for KK2.0. Mean EPG between KK2.0 and KK1.0 were not statistically different across STH species and timepoints, except for T. trichiura at 24-hours (higher mean EPG for KK1.0, p = 0.036). When considering infection intensity levels, KK2.0 identified 10% more of the total population as low-infection intensity samples of A. lumbricoides than KK1.0 (p ≤ 0.001, both timepoints) and similar to KK1.0 for T. trichiura and hookworms. Varying agreement existed between KK1.0 and KK2.0 in detecting STH eggs (A. lumbricoides: moderate; T. trichiura: substantial; hookworms: slight). However, these findings should be interpreted carefully as there are certain limitations that may have impacted the results of this study. This study demonstrates the potential of the AI-DP-based method for STH diagnosis. While similar to KK1.0, the AI-DP-based method outperforms it in certain aspects. These findings underscore the potential of advancing the AI-DP KK2.0 prototype for dependable STH diagnosis and furthering the development of automated digital micr