In a study published in Frontiers Bioscience-Landmark, a research group led by Prof. LI Bei from the Changchun Institute of Optics, Fin Mechanics and Physics (CIOMP) of the Chinese Academy of Sciences (CAS), in collaboration with Wenzhou Medical University, proposed a detection and analysis method of bacterial Raman spectroscopy based on image stitching and automatic identification algorithm can be used for rapid, accurate and fully automated of analysis of the Raman spectrum of all bacteria at high magnification with multiple fields of view.

Due to antibiotic abuse, the problem of bacterial resistance is becoming increasingly serious. Raman spectroscopy enables rapid, reproducible, and non-invasive qualitative and quantitative analysis of bacteria resistance. However, When the Raman spectra of a large number of bacteria need to be collected, the traditional method can only scan a single field of view to identify bacteria, and it is necessary to manually select the bacteria to be tested. If there are too many fields of view, it will increase the volume of the experiment, the efficiency will be reduced.

The purpose of image stitching is to automatically calculate the number of fields of view needed by scanning small fields of view with the number of bacteria entered by the user, and to stitch these overlapping small field of view images into one multiple fields of view image to obtain a panoramic image with enough bacteria. Image stitching includes image identification, image registration, global optimization, and image blending.

In order to identify and locate single cells and excluded the impurities and cluster phenomena, the input images are preprocessed first, and the target cells are filtered according to the characteristics of area, roundness, eccentricity, and convexity. We number the identified bacteria to quickly and accurately locate them and provide assistance in verifying their Raman spectra if necessary.

The coordinates of center points of the single cells were communicated to the motorized stage to get the Raman Spectra of bacteria. The Savitzky-Golay (SG) filtering method is used to wight filter the Raman Spectrum, an adaptive iteratively reweighted Penalized Least Squares (airPLS) algorithm is used to calibrate the baseline of the smoothed Raman Spectrum. Automatic peaking of Raman spectra using Persistence1D to extract, pair, and sorted local minima and local maxima according to their persistence. Then antibiotic resistance is analyzed.

This study can be used as a high-throughput, rapid and accurate method to detect antimicrobial resistance by Raman spectroscopy for researching the persistence and spread of antibiotics in bacterial pathogens.