PS4-18. Proof-of-concept Evaluation of Multi-frequency Bioelectrical Impedance Analysis for Non-invasive Detection of Caseous Lymphadenitis in Meat Goats.

Abstract

Caseous lymphadenitis (CL) is a chronic bacterial disease that negatively impacts meat goat productivity, carcass value, and herd biosecurity. Existing detection methods depend on visual inspection or laboratory confirmation, which may fail to identify early physiological changes. This proof-of-concept study assessed whether multi-frequency bioelectrical impedance analysis (BIA) can detect measurable tissue-level differences between CL-affected and healthy meat goats through group-aware machine learning and statistical modeling. A total of 2,701 impedance measurements were obtained from 75 male Spanish goats at three frequencies (50, 100, and 180 kHz), comprising 971 CL and 1,730 healthy observations. Derived impedance features included resistance (Rs), reactance (Xc), impedance magnitude (Z), reactance ratio, and frequency slopes. To avoid animal-level data leakage, classification models were validated using 5-fold Group KFold cross-validation. Among the evaluated models, the Support Vector Machine demonstrated the highest discriminatory performance (AUC = 0.61), surpassing ElasticNet (AUC = 0.49) and XGBoost (AUC = 0.44). At the optimized Youden threshold (0.035), sensitivity reached 1.00, indicating complete detection of CL cases, whereas specificity was 0.14. Despite modest discrimination, the high sensitivity supports the potential application of BIA as an early screening tool rather than a standalone diagnostic method. Mixed-effects regression modeling (REML; 75 animal groups) identified reactance (Xc) as a statistically significant predictor of CL status (p = 0.032), whereas resistance was not significant (p = 0.957). Principal component analysis demonstrated measurable physiological divergence between groups. SHAP interpretability analysis indicated that phase angle at 180 kHz and reactance-related features contributed most to classification performance. Learning curve assessment indicated stable signal extraction across animals, supporting the reproducibility of impedance-derived features. Decision curve analysis suggested a potential net benefit at low-risk screening thresholds. As a proof-of-concept investigation, this study demonstrated that multi-frequency BIA can capture biologically meaningful physiological signatures associated with CL in meat goats. These findings provide foundational evidence for integrating impedance-based sensing into multi-modal precision livestock health monitoring systems.