Classification of Mycoplasma Pneumoniae Strains Based on
One-Dimensional Convolutional Neural Network and
Raman Spectroscopy
ZHAO Yong1, HE Men-yuan1, WANG Bo-lin2, ZHAO Rong2, MENG Zong1*
1. School of Electrical Engineering, Yanshan University, The Key Laboratory of Measurement Technology and Instruments of Hebei Province, Qinhuangdao 066004, China
2. School of Information Science and Engineering, Yanshan University, The Key Laboratory for Special Fiber and Fiber Sensor of Hebei Province, Qinhuangdao 066004, China
Abstract：Mycoplasma pneumoniae is the main cause of human respiratory diseases. Clinically, the symptoms of patients infected with different mycoplasma pneumoniae are very similar, so it is difficult to distinguish the type of mycoplasma pneumoniae according to the symptoms and give medication. Therefore, the accurate identification of mycoplasma pneumoniae strain type is of great significance for the pathogenesis and epidemiological research of the disease, and accurate clinical treatment. Raman spectrum has been paid more and more attention because of its advantages of fast-speed, high efficiency, pollution-free and non-destructive analysis. One-dimensional Convolution Neural Network (1D-CNN) is a kind of pre-feedback network with a deep structure, including Convolution operation. It has been successfully applied in the analysis of speech and vibration signals. The combination of the One-Dimensional Convolution Neural Network and the Raman spectral data of the main genotypes of mycoplasma pneumoniae M129 and FH were used as the research objects to realize mycoplasma classification pneumoniae strains. The spectral data enhancement method expands the original spectral data set, and the one-dimensional convolution neural network model was trained, and the problem of data hunger of convolutional neural network caused by small samples was solved. In order to obtain the best classification effect of mycoplasma pneumoniae and accelerate the learning process, the model structure was optimized, and the best model parameters were determined. Gaussian noise, Poisson noise and multiplicative noise are often mixed in Raman spectral measurement. Gaussian noise, Poisson noise and multiplicative noise are often mixed in Raman spectral measurement. In order to optimize the anti-noise ability of the model, Gaussian noise, Poisson noise and multiplicative noise were superimposed on the original spectrum respectively, and the 1D-CNN model was trained and compared with the models built by traditional algorithms such as LDA, KNN and SVM. The experimental results show that for the Raman spectra superimposed with Gaussian noise, Poisson noise and multiplicative noise, the classification accuracy of the models based on 1D-CNN method has achieved 98.0%, 97.0% and 97.0%, respectively, which are all much higher than those of the models based on LDA, KNN and SVM algorithms. At the same time, the 1D-CNN model can achieve 92.5% classification accuracy when the noise reacheds the 55 dBW interference factor, aiming at the noise with different intensities of 5, 15, 25, 35, 45 and 55 dBW.Therefore, it is feasible to apply a one-dimensional convolutional neural network combined with Raman spectrum technology to the classification of mycoplasma pneumoniae strain types, which has the advantages of strong anti-noise ability and high classification accuracy. This study provides a new idea for the rapid diagnosis of mycoplasma pneumoniae pneumonia.
[1] Waites K B, Xiao L, Liu Y, et al. Clinical Microbiology Reviews, 2017, 30(3): 747.
[2] He J, Liu M H, Ye Z F, et al. Molecular Medicine Reports, 2016, 14(5): 4030.
[3] Parrott G L, Kinjo T, Fujita J. Frontiers in Microbiology, 2016, 7: 513.
[4] Loens K, Leven M. Frontiers in Microbiology, 2016, 7: 448.
[5] Miyashita N, Kawai Y, Kato T, et al. Journal of Infection and Chemotherapy, 2016, 22(5): 327.
[6] Sano G, Itagaki T, Ishiwada N, et al. Journal of Medical Microbiology, 2016, 65(10): 1105.
[7] Khan S, Ullah R, Khan A, et al. Photodiagnosis and Photodynamic Therapy, 2018, 23(9): 89.
[8] Chen C, Yang L, Zhao J, et al. Optik, 2020, 203: 164043.
[9] Liu J, Osadchy M, Ashton L, et al. Analyst, 2017, 142: 4067.
[10] Shao X, Zhang H, Wang Y, et al. Nanomedicine: Nanotechnology, Biology and Medicine, 2020, 29: 102245.
[11] LI Qing-xu, WANG Qiao-hua, GU Wei, et al(李庆旭, 王巧华, 顾 伟, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2020, 40(12): 3847.
[12] Henderson K C, Benitez A J, Ratliff A E, et al. PLOS ONE, 2015, 10(6): e0131831.
