Automatic quantitative analysis of metabolism inactivation concentration in single bacterium using stimulated Raman scattering microscopy with deep learning image segmentation

Rapid antimicrobial susceptibility testing (AST) is urgently needed to slow down the emergence of antibiotic-resistant bacteria and treat infections with correct antibiotics. Stimulated Raman scattering (SRS) microscopy is a technique that enables rapid chemical-bond imaging with sub-cellular resolution. It can obtain the AST results with a single bacterium resolution. Although the SRS imaging assay is relatively fast, taking less than 2 h, the calculation of single-cell metabolism inactivation concentration (SC-MIC) is performed manually and takes a long time. The bottleneck tasks that hinder the SC-MIC throughput include bacterial segmentation and intensity thresholding. To address these issues, we devised a hybrid algorithm to segment single bacteria from SRS images with automatic thresholding. Our proposed method comprises a U-Net convolutional neural network (CNN), DropBlock, and secondary segmentation post-processing. Our results show that SC-MIC calculation can be accomplished within 1 min and more accurate segmentation results using deep learning-based bacterial segmentation method, which is essential for its clinical applications.