Semi-Automated System Detects Cleaner Fish Interactions With 90% Accuracy but 15% False Positive Rate

A research team developed a semi-automated system that tracks and classifies cleaning interactions between the cleaner wrasse Labroides dimidiatus and the powder blue tang Acanthurus leucosternon. The system achieved 90% accuracy in detecting interactions while reducing the amount of footage requiring manual annotation by 75%. 1038/s41598-026-56200-6.

Authors Raul Oliveira, Nuno Cruz Garcia, and José Ricardo Paula conducted the work. Raul Oliveira and José Ricardo Paula are affiliated with MARE – Marine and Environmental Sciences Centre & ARNET – Aquatic Research Network, Laboratório Marítimo da Guia, Faculdade de Ciências Universidade de Lisboa.

Nuno Cruz Garcia is affiliated with LASIGE, Departamento de Informática, Faculdade de Ciências Universidade de Lisboa.

José Ricardo Paula is the corresponding author. The experiments took place in a controlled three-dimensional laboratory setting.

Researchers used DeepLabCut for markerless pose estimation to track both fish species simultaneously. The model tracked both individuals with low error rates. From the tracking data, the team built a classification algorithm that detected cleaning interactions with 90% accuracy.

The algorithm misclassified approximately 15% of non-interactions as interactions and identified 25% of video content as containing interactions. Cleaner fish engage in mutualistic interactions by removing ectoparasites from client species, a behaviour that has traditionally been quantified through labour-intensive manual video analysis.

The project received support from a fellowship from the “la Caixa” Foundation (ID 100010434) with fellowship code LCF/BQ/PR24/12050006.

The work was supported by FLAD Science Award Atlantic—AtlanticDiversa (Proj. 2026/0095) funded by FLAD—Fundação Luso-Americana para o Desenvolvimento. 54499/PTDC/BIA-BMA/0080/2021) to JRP.

54499/UID/04292/2025). 54499/UID/PRR/04292/2025). The work was supported by the EQUIPAR+2 programme attributed to MARE (Reference: UID/PRR2/04292/2025).

54499/LA/P/0069/2020). The authors declare no competing interests. 0 International License.

Com reported the study details.