Portable near-infrared spectroscopy combined with machine learning for rapid detection of broiler breast myopathies

Growth-related breast myopathies are among the major meat quality defects encountered in modern broiler production, affecting breast meat quality and requiring affected fillets to be diverted to further processing. This study aimed to evaluate the feasibility of a pocket-sized near‑infrared (NIR) spectrometer for detecting white striping (WS), wooden breast (WB), and spaghetti meat (SM) in broiler breast meat. A total of 4579 broilers from a commercial crossbred and a broiler breeders line were sampled across 8 flocks and multiple slaughter days, and fillets were categorized by a trained evaluator for WS, WB, and SM severity. Breast surfaces were scanned five times at standardized anatomical locations using a portable spectrometer (740-1070 nm), and the spectra were averaged prior to chemometric analyses. After data cleaning, 4500 samples were retained, and partial least squares discriminant analysis (PLS‑DA) models (70% training set; 30% testing set) were developed and evaluated using balanced accuracy (BA), sensitivity, and specificity. Results revealed models that BA in the testing ranged from 50 to 78%, depending on the classification scenario. Models for SM and isolated WS or WB showed limited discrimination (≤56%), reflecting class imbalance and spectral overlap with concurrent myopathies. Intermediate discrimination was obtained for Non-defect (ND) vs WS and/or WB (75%), ND vs any defect (75%), WS severity (74%), and WB severity (73-74%). The highest BA values were achieved for models contrasting ND fillets with WS-associated myopathies (78%) or WB-associated myopathies (77%). For the ND vs any defect model, sensitivity and specificity were 67% and 82%, respectively, indicating a higher ability to identify ND fillets than defective fillets. Variable‑importance analysis showed defect‑specific spectral regions: 800-900 nm for SM (water‑related absorptions), 900-950 nm for WS (lipid‑related features), and 1000-1070 nm for WB (protein/collagen‑associated absorptions), reflecting underlying biochemical alterations. In conclusion, a pocket‑sized NIR spectrometer can serve as a practical, rapid first‑pass screening tool for detecting combined broiler breast myopathies at processing lines, although isolated and low‑prevalence defects remain challenging to classify reliably.