Grade classification of human glioma using a convolutional neural network based on mid‐infrared spectroscopy mapping

Abstract

This study proposes a convolutional neural network (CNN)‐based computer‐aided diagnosis (CAD) system for the grade classification of human glioma by using mid‐infrared (MIR) spectroscopic mappings. Through data augmentation of pixels recombination, the mappings in the training set increased almost 161 times relative to the original mappings. The pixels of the recombined mappings in the training set came from all of the one‐dimensional (1D) vibrational spectroscopy of 62 (almost 80% of all 77 patients) patients at specific bands. Compared with the performance of the CNN‐CAD system based on the 1D vibrational spectroscopy, we found that the mean diagnostic accuracy of the recombined MIR spectroscopic mappings at peaks of 2917 cm−1, 1539 cm−1 and 1234 cm−1 on the test set performed higher and the model also had more stable patterns. This research demonstrates that two‐dimensional MIR mapping at a single frequency can be used by the CNN‐CAD system for diagnosis and the research also gives a prompt that the mapping collection process can be replaced by a single‐frequency IR imaging system, which is cheaper and more portable than a Fourier transform infrared microscopy and thus may be widely utilized in hospitals to provide meaningful assistance for pathologists in clinics.

Fast fingerprint scanning and chemical composition identification of a large-scale material surface by a combined HSI-LIBS operation

“…Spectroscopy technology provides a simple, fast and micro-destructive approach to detecting material surface structure, morphology, and chemical composition information, making it a research hotspot. It includes laser-induced breakdown spectroscopy (LIBS), 2 Raman spectroscopy (RS), 3 infrared spectroscopy, 4 hyperspectral imaging (HSI), 5 optical imaging, 6 and other techniques. In recent years, researchers have therefore attempted to come up with more efficient methods to achieve the chemical composition and distribution of a material surface using a point-to-point scan method such as LIBS and RS mapping.…”