Machine learning-enhanced confocal Raman imaging enables label-free diagnosis and spatial metabolic profiling of isoniazid-induced hepatotoxicity.

Abstract

Rationale: Isoniazid-induced liver injury (INH-ILI) poses a significant clinical challenge due to the lack of reliable, non-invasive, and real-time diagnostic tools. Here, we present an integrated platform that combines label-free confocal Raman spectroscopy imaging, machine learning (ML), and targeted metabolomics to identify and classify INH-ILI in a murine model. Methods: An INH-ILI mouse model was established, and Raman imaging and subsequent data analysis were performed on the control and INH-ILI at 7, 14, 21, and 28-day groups. Alterations in hepatic metabolites following INH-ILI were elucidated. Furthermore, ML techniques were employed to identify subtle differences between the control and INH-ILI groups. Results: Distinct Raman spectral shifts, notably the emergence of a 1638 cm^-1 peak in injured liver tissues compared to characteristic peaks at 1203, 1266, and 1746 cm^-1 in controls, were observed. ML models including support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and convolutional neural network (CNN) have achieved accurate staging and classification of INH-ILI (AUC > 0.95). Metabolomic analysis further confirmed disruptions in lipid and aromatic amino acid metabolism, particularly involving phenylalanine-tyrosine imbalance linked to oxidative stress. Conclusions: This method enables precise, high-throughput, and spatially resolved diagnosis of INH-ILI, with strong potential for clinical translation in drug-induced liver injury assessment.