Lung Cancer Diagnosis Model Based On High Energy Ultraviolet Laser-Mass Spectrometry And Deep Learning

Background:There are various small molecular substances in the human bloodstream,including amino acids,peptides,fatty acids,nucleic acids,and other small molecular components.During the development and progression of tumors,the number,species and structure will undergo various changes.The study of small molecular substances is of great significance to explore lung cancer’s occurrence and development.Compared with traditional physical and chemical methods,mass spectrometry has high-throughput advantages,which can rapidly detect small molecular substances.Mass spectrometry is a newly developed biological image detection technology that can simultaneously detect the mass and spatial distribution information of various molecules without labeling samples and realize high-throughput visualization of molecules on samples’ surfaces.One of the bottlenecks limiting mass spectrometry detection is the low ionization rate of samples,especially for samples with complex components such as serum.In order to improve the ionization rate of sample detection,‘the comprehensive experimental research device based on adjustable ultraviolet coherent light source(Dalian coherent light source for short)’ and Matrix-assisted laser desorption/ionization(MALDI)are used to detect serum samples.Dalian coherent light source is jointly developed by the Dalian Institute of Chemical Physics and Chinese Academy of Sciences,and Shanghai Institute of Applied Physics.It emits a high-energy ultraviolet laser with excellent properties such as ultra-high brightness,ultra-short pulse,and full coherence.The combination of high-energy ultraviolet laser and MALDI can significantly improve the ionization rate and detection rate of samples,shorten the experimental research time and improve the spatial resolution.Due to serum samples’ complex characteristics,there are many problems in the detected mass spectra,such as large quantity,high complexity,and low significant correlation.Traditional mass spectrometry cannot meet the demand for scientific research.Artificial intelligence is a deep learning technology that has been widely used in the field of health biology.Because of its inherent characteristics,it has natural advantages in dealing with complex and massive data.At present,deep learning technology has invented numerous functional artificial intelligence models by analyzing and learning a large number of imaging data,histopathological data and bioengineering information,which has made significant contributions to diagnosis of a disease,evaluation of curative effect,adjuvant therapy and drug prediction.Method:This study aims to utilize high-energy ultraviolet laser-mass spectrometry technology to detect the serum of lung cancer patients and healthy people,obtain the serum mass spectrum of lung cancer and healthy people,and then use the artificial intelligence deep learning technology to analyze and train the mass spectrum,and construct the artificial intelligence diagnosis and treatment model of lung cancer based on the serum mass spectrum.The deep learning model of the artificial intelligence used in this article is a residual network(Res Net for short),a convolution network based on a basic residual structure.The application of a residual network dramatically improves the degradation of traditional convolution network caused by the increase of training depth.Firstly,we preprocess the data in order to normalize and dematerialize the data.Secondly,the depth residual network model is constructed.The model’s contents are as follow :(1)Setting training objectives and optimizing methods;(2)Constructing a convolution layer and a residual layer;(3)Cross-validation of mass spectrometry data;(4)Sensitivity analysis was carried out on the characteristic spectral peaks;(5)Balance accuracy is introduced to judge the final network.Conclusion:The evaluation standard of deep learning lung cancer diagnosis model is the accuracy of verifying whether an unknown map is lung cancer.Accuracy was 60% to70% using gold nanocluster matrix and 55% to 65% using traditional DHB matrix.In the map of gold nanocluster matrix,after removing the influence of matrix,the verification accuracy can be stabilized at 65%-70%.In the screening of potential lung cancer markers,it was found that there were both lung cancer and healthy group,but there were 3 more significant and specific mass spectrum peaks in lung cancer group.There were 10 unique peaks of mass spectrum in lung cancer group.Among them,there were 3 mass spectrum peaks with potential as diagnostic markers for lung cancer.