Detection of Carbapenem Resistance in Klebsiella pneumoniae Using Vision Transformers and MALDI-TOF Proteomic Profiles

Abstract

Antimicrobial resistance constitutes an escalating global health concern, significantly increasing morbidity, mortality, and healthcare-associated costs. Rapid and accurate detection of resistant bacterial strains is crucial for optimizing clinical decision-making and mitigating their dissemination. This study investigates the integration of artificial intelligence (AI) and mass spectrometry techniques for the classification of carbapenem-resistant Klebsiella pneumoniae strains. Specifically, we leverage Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) data and a Convolutional Vision Transformer (CvT) model to enhance classification accuracy. A dataset comprising 180 proteomic spectra—100 from carbapenem-sensitive isolates and 80 from resistant isolates—was analyzed. The proposed CvT model demonstrated superior performance compared to traditional machine learning and deep learning architectures, achieving the highest mean accuracy of 80.61% and a robust F1-score of 80.19% across a five-fold cross-validation. Grad-CAM visualizations further enhanced the model’s interpretability by identifying critical proteomic patterns associated with resistance. The findings underscore the potential of vision transformers for detecting antimicrobial resistance, offering a novel approach to optimize microbiological diagnostics.