关键词: 木质化鸡胸肉；肉质分级；机器学习；挥发性风味物质；沙普利加和解释；Mantel相关性分析

Abstract: This study aimed to address the growing prevalence of wooden breast (WB) in fast-growing broilers and the limitations of conventional detection methods, which are often subjective and lack precision in grading. Normal breast (NB), mild wooden breast (LWB), moderate wooden breast (MWB), and severe wooden breast (SWB) from Cobb broilers were evaluated for quality parameters including pH, color, water-holding capacity, textural properties, and shear force; additionally, volatile compounds were analyzed to develop a comprehensive, multidimensional grading system for wooden breast meat. Headspace gas chromatography-ion mobility spectrometry (HS-GC-IMS) identified a total of 45 volatile compounds, among which 16 were determined as key differential substances. The contents of aldehydes and esters exhibited a significant decreasing trend with increasing WB severity. Principal component analysis (PCA) indicated that texture parameters and water-holding capacity were critical indicators for differentiating WB grades. Based on these findings, a backpropagation artificial neural network (BP-ANN) model was developed, demonstrating high classification accuracy with training and testing set accuracies of 98.81% and 94.44%, respectively. Shapley additive explanations (SHAP) analysis further identified resilience, chewiness, pH, drip loss, and L* value as key discriminant indicators. Mantel tests showed a significant positive correlation between resilience and 1-propanol, between chewiness and 1-octenal, and between L* value and 1-octenal. These findings suggest that structural damage of muscle fibers and enhanced lipid oxidation during WB development may influence the formation of volatile flavor compounds. This study contributes to the theoretical understanding of the multidimensional mechanisms underlying meat quality deterioration.

Key words: wooden breast; meat quality grading; machine learning; volatile flavor substances; SHapley Additive exPlanations; Mantel correlation analysis