A deep learning-based automated image analysis for histological evaluation of broiler pectoral muscle

Jonathan Dayan; Noam Goldman; Daniel Waiger; Tal Melkman-Zehavi; Orna Halevy; Zehava Uni

doi:10.1016/j.psj.2023.102792

A deep learning-based automated image analysis for histological evaluation of broiler pectoral muscle

Poult Sci. 2023 Aug;102(8):102792. doi: 10.1016/j.psj.2023.102792. Epub 2023 May 19.

Authors

Jonathan Dayan¹, Noam Goldman², Daniel Waiger³, Tal Melkman-Zehavi¹, Orna Halevy¹, Zehava Uni⁴

Affiliations

¹ Department of Animal Science, The Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem, Rehovot 7610001, Israel.
² Koret School of Veterinary Medicine, The Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem, Rehovot 7610001, Israel.
³ Center for Scientific Imaging, The Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem, Rehovot 7610001, Israel.
⁴ Department of Animal Science, The Robert H. Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem, Rehovot 7610001, Israel. Electronic address: zehava.uni@mail.huji.ac.il.

Abstract

Global market demand for chicken breast muscle with high yield and quality, together with the high incidence rate of breast muscle abnormalities in recent years highlights the need for tools that can provide a rapid and precise evaluation of breast muscle development and morphology. In this study, we used a novel deep learning-based automated image analysis workflow combining Fiji (ImageJ) with Cellpose and MorphoLibJ plugins to generate an automated diameter and cross-sectional area quantification for broiler breast muscle. We compared data of myofiber diameter from 14-day-old broiler chicks, generated either by manual analysis or by automated analysis. Comparison between manual and automated analysis methods exhibited a striking accuracy rate of up to 99.91%. Moreover, the automated analysis method was much faster. When the automated analysis method was implemented on 84 breast muscle cross-section images it characterized 59,128 myofibers within 4.2 h, while manual analysis of 27 breast muscle cross-section images enabled analysis of 17,333 myofibers in 54 h. The automated image analysis method was also more productive, producing data sets of both diameter and cross-sectional area at an 80-fold higher rate than the manual analysis (26,279 vs. 321 data sets per hour, respectively). In order to demonstrate the ability of this automated image analysis tool to detect differences in breast muscle histomorphology, we applied it on cross sections from chicks of control and in ovo feeding group, injected with a methionine source [2-hydroxy-4-(methylthio) butanoic calcium salt (HMTBa)], known to effect skeletal muscle histomorphology. Analysis was performed on 19,807 myofibers from the control group and 21,755 myofibers from the HMTBa group and was completed in less than 1 h. The clear advantages of this automated image analysis workflow characterized by high precision, high speed, and high productiveness demonstrate its potential to be implemented as a reproducible and readily adaptable research or diagnostic tool for chicken breast muscle development and morphology.

Keywords: automated image analysis; breast muscle; broiler; chicken; histology.

MeSH terms

Animals
Chickens / physiology
Deep Learning*
Image Processing, Computer-Assisted / methods
Methionine
Muscle, Skeletal
Pectoralis Muscles*

Substances

Methionine