# Automated Detection of Parasitic Elements in Veterinary Fecal Samples Using a Deep Learning-Based Object Detection Framework

## TL;DR

This paper presents an automated system using deep learning to detect six types of parasites in veterinary fecal samples, improving diagnostic speed and accuracy.

## Contribution

The study introduces a YOLOv8-based framework for automated detection of six parasitic elements in fecal microscopy images with high accuracy and speed.

## Key findings

- The YOLOv8n model achieved a mean average precision of 0.982 across five parasite types.
- Detection speeds averaged under 60 milliseconds per image on a standard CPU.
- 2500× magnification provided the best balance for routine screening, while 10,000× was useful for detailed confirmation.

## Abstract

Microscopic examination of fecal samples is a fundamental method for diagnosing parasitic infections in dogs and cats, but traditional manual inspection is labor-intensive, time-consuming, and requires specialized expertise that may not always be available in busy veterinary practices. This study developed an automated detection system using a deep learning model called YOLOv8n to identify six types of parasites commonly encountered in veterinary medicine: egg-producing parasites (Spirometra, Dipylidium, hookworm, Ascaris) and single-celled parasites (Giardia cysts and Trichomonas trophozoites). Images were captured at three different magnifications (1000×, 2500×, and 10,000×) to evaluate detection performance across varying levels of morphological detail and field of view. The dataset comprised 326 microscopic images containing over 3700 individually annotated parasitic elements. The YOLOv8 model achieved high accuracy for five parasite types and demonstrated robust performance even for the smallest and most challenging targets, with detection speeds under 60 milliseconds per image on a standard computer. The 2500× magnification was identified as the most practical choice for routine screening, offering an optimal balance between fine detail and efficient sample coverage. The 10,000× magnification proved valuable for confirming ambiguous cases requiring detailed morphological examination. This automated tool has the potential to assist veterinarians by reducing manual screening time, improving diagnostic consistency, and ultimately supporting better healthcare outcomes for companion animals through faster and more reliable parasite detection.

Parasitic infections in veterinary medicine are commonly diagnosed through microscopic examination of fecal samples, yet traditional manual methods are labor-intensive and subject to diagnostic variability. This study investigates YOLOv8 for automated identification of parasitic elements in fecal microscopy images. Six parasitic taxa were analyzed at 1000×, 2500×, and 10,000× magnifications: Spirometra eggs, Dipylidium egg packets, hookworm eggs, Ascaris eggs, Giardia cysts, and Trichomonas trophozoites. The dataset comprised 326 images with 3710 annotated objects, split at the sample level into training (70%), validation (15%), and testing (15%) sets. The YOLOv8n model achieved mean average precision (mAP@0.5) of 0.982 ± 0.015 across 5-fold cross-validation. Per-class AP exceeded 0.97 for five taxa, with Trichomonas achieving 0.952. Inference time averaged under 60 ms per image on a standard CPU. These results demonstrate that YOLOv8 provides accurate and efficient detection of diverse parasitic elements, supporting its potential as a clinical screening tool.