Automated detection of Eimeria tenella from hematoxylin and eosin-stained chicken cecal tissues using YOLOv4-based deep learning: A proof-of-concept study.

Avian coccidiosis is a disease primarily characterized by diarrhea caused by intestinal infection with protozoa of the genus Eimeria, causing significant economic impact worldwide. Early diagnosis of infected chickens is crucial for effective control of this disease. While several studies have reported the usefulness of deep learning algorithms for recognizing oocysts on smear specimens, to the authors' knowledge, no reports have examined this approach on histopathological specimens. In this study, based on the hypothesis that deep learning can assist histopathological diagnosis, we applied an object detection algorithm to demonstrate models to automatically detect schizonts and macrogametocytes of E. tenella from histopathological images as a proof-of-concept. We prepared 960 image patches from hematoxylin and eosin-stained cecal specimens of four chickens orally infected with E. tenella. These were divided into training and validation data, and models were developed using the You Only Look Once version 4 (YOLOv4) and YOLOv4‑tiny algorithms. The resulting models were then used to detect parasites in tissue images from field cases. The resulting mean average precisions were 78.08-78.87% for YOLOv4 and 64.26-68.13% for YOLOv4‑tiny. Although this study has limitations in the dataset size and may not fully capture the real-world variability, it suggests that YOLOv4 is feasible for detecting coccidia in histopathological specimens. The proposed approach has the potential to support histopathological diagnosis and facilitate the detection of parasitic stages during routine pathological examinations, contributing to more objective and reproducible pathological evaluations.