ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Unsupervised Discovery of Biologically Interpretable Fecal Pathology Phenotypes for Objective Coccidiosis Severity Assessment in Broilers

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600632.(doi:10.13031/aim.202600632)
Authors:   Oluwadamilola Moyin Oso, Hosna Mohammadilalabadi, Guoming Li, Lorraine A Fuller, Samuel E Aggrey
Keywords:   Broilers; Coccidiosis, Fecal phenotyping, Precision livestock farming, Unsupervised clustering.

Abstract. Accurate fecal scoring remains a major challenge in poultry health, as human scoring relies heavily on subjective visual clues. Developing objective digital fecal scoring systems requires reproducible phenotypic definitions; however, it is critical to first identify the natural fecal pathology phenotypes. This study aimed to develop an unsupervised biologically informed framework for discovering fecal pathology phenotypes to support infection severity characterization. A total of 240 fecal samples were collected from non-infected and coccidiosis-infected broilers (3 to 6 days post infection) at days 13-16th of age. A total of 83 features that were biologically relevant to clinical pathology were extracted from 234 valid samples. The features were organized into 4 biological blocks. Weighted early fusion and principal component analysis reduced feature dimensions for K-means clustering and hierarchical sub-clustering. The result showed two optimum clusters (k=2, silhouette score=0.231). Hierarchical sub clustering reveals 2 phenotypic strata within cluster 0 (k=2, silhouette score = 0.25) and 3 phenotypic strata within cluster 1 (k=3, silhouette score = 0.61). Post hoc analysis (both Kruskal–Wallis and Spearman tests) showed significant association (p < 0.001) between human scores and the clusters. Cluster stability analysis shows low clustering stability (mean adjusted rand index (ARI) = 0.06) for only top-level clusters, whereas joint cluster-subcluster hierarchies showed substantially high stability (mean ARI =0.54). This proposed framework provides a foundation for objective and scalable digital fecal severity assessment in poultry health by utilizing biologically understandable features.

(Download PDF)    (Export to EndNotes)