High-accuracy Rapid Identification and Classification of Mixed Bacteria Using Hyperspectral Transmission Microscopic Imaging and Machine Learning
He Zhu, Jing Luo, Jiaqi Liao, Sailing He
Abstract
In this paper, we developed a hyperspectral transmission microscopic imaging (HTMI) system for rapid detection of pathogenic bacteria, which can realize precise identification and classification of mixed pathogenic bacteria to a single-bacterium level. The system works in transillumination patterns and a self-developed dispersive hyperspectral imaging module is used as the detection setup, providing spectral images with high SNR, and showing excellent performances with spatial resolution of 2.19 m and spectral resolutions less than 1 nm. Hyperspectral microscopic imaging of five types of bacteria in low concentration were performed. The merging spatial-spectral profiles of individual bacteria for each species were extracted and utilized for species identification, achieving high classification accuracy of 93.6% using a simple PCA-SVM method. Species identification experiments of the mixed bacterial samples were further carried out, and the results demonstrate the validity and capability of the system assisted with simple machine learning methods to be used as an effective and rapid diagnostic tool for elaborate identification of mixed bacterial pathogen samples, providing guidance for the use of correct antibiotics.