Publikationsdatenbank

Automated scoring of exercise-induced pulmonary hemorrhage (EIPH) in equine bronchoalveolar lavage cytology (2019)

Art

Vortrag

Autoren

Bertram, C. A. (WE 12)

Mahrzahl, C.

Aubreville, M.

Barton, A. K. (WE 17)

Maier, A.

Hill, J.

Klopfleisch, R. (WE 12)

Kongress

Joint Congress of Veterinary Pathology and Veterinary Clinical Pathology

Arnheim, Niederlande, 25. – 28.09.2019

Quelle

Veterinary clinical pathology

Bandzählung: 48

Heftzählung: 4

Seiten: 797

ISSN: 0275-6382

Sprache

Englisch

Verweise

URL (Volltext): https://onlinelibrary.wiley.com/doi/epdf/10.1111/vcp.12798

DOI: 10.1111/vcp.12798

Kontakt

Institut für Tierpathologie

Robert-von-Ostertag-Str. 15
14163 Berlin
+49 30 838 62450
pathologie@vetmed.fu-berlin.de

Abstract / Zusammenfassung

Background:
Exercise-induced pulmonary hemorrhage (EIPH) is a common syndrome in sport horses with a negative impact on per-formance. Cytology of bronchoalveolar lavage fluid (BALF) encom-passing the use of a scoring system is considered the most sensitive diagnostic method. Macrophages are classified into five grades de-pending on the degree of cytoplasmic hemosiderin content. Manual grading is not only monotonous and time-consuming, but also prone to inter- intra-observer variability.

Objective:
To evaluate feasibility of an automated EIPH scoring.

Methods:
Cytocentrifugation of 17 equine BALF with variable dis-ease extent were obtained during routine diagnostic service and stained for iron content. All macrophages in whole slide images (WSI) were labeled and graded according to published methods (ground truth). Automated cell detection and classification was developed by a state-of-the-art deep learning pipeline based on RetinaNet. Some memory and execution speed modifications were introduced in order to generate analysis of the entire WSI in less than 2 minutes. Concordance of the manual (repeated classification) and algorithmic approach was determined in 2000 random macrophages.

Results:
Concordance with the ground truth data was 81.1% for the manual classification and 85.4% for the computerized classification with just one of the 2000 cells being discordant by more than one tier. The object detection approach has a high mean average preci-sion of 0.64 (σ = 0.10, intersection over union threshold = 0.5) over the WSI with a mean grading score error of 0.07 (σ = 0.06).Conclusion: Automated cell detection and classification is a prom-ising method for accurate, reproducible, and quick EIPH scoring in WSI.