Coordinate Transformer Network for Prediction of Pseudomonas Aeruginosa’s Drug Resistance

Pseudomonas aeruginosa (PA) is a pathogenic bacterium in hospital infections. In recent years, due to the abuse of antibiotics, the number of multi-drug resistant PA (MDRPA) increases year by year. These resistant pathogens pose a challenge to clinicians in terms of rapid detection of bacterial resistance and rational use of antibiotics. To address it, the coordinate transformer network (CTN) is proposed to predict drug resistance of PA from fluorescent images. Specifically, we explore ResNeSt-50 as the backbone. The coordinate attention (CA) embeds location information into channel attention so that the target region can be accurately captured. Because of vision transformer’s impressive global modeling ability, we incorporate a hierarchical multi-head self-attention (H-MHSA) module to the network, which take both advantages of deep convolutional neural network and transformer. The experimental results on our self-collected data show that our CTN is promising in identifying sensitive and multi-drug resistant PA.