Application Of Fluorescence Spectrum And Tumor Markers Combined With Artificial Neural Network In Diagnosis Of Lung Cancer

Background and ObjectiveLung cancer is one of the serious diseases that threaten human life and health,In western developed countries,the morbility and mortality of lung cancer keep the first place in all kinds of cancers.In China,the mortality of lung cancer keeps the first place in cities,second in countrysides,respectively.Lung cancer rapidly develops and aggravates and the incidence is high.The patients who have the clinical symptom would lose the optimal treatment opportunities,prognosis is not satisfying.Therefore, early detection,early diagnosis and early treatment are the keys to increase the survival opportunity and improve prognosis of patients with lung cancer.At present, imaging technology,tissue cytology technology and bronchofibroscope technology are the main technologies in the diagnosis of lung cancer.However,all of the technologies have their owner disadvantages.Serology detection(including fluorescence spectrum and tumor markers) was highly thought as convenience and hypersensitivity by domestic and foreign researchers,it has become one of the most significant ways in the early detection and diagnosis of lung cancer.Serum fluorescence spectrum can reflect the variations of component,quantity and microenvironment of fluorescent material in serum,therefore,serum fluorescence spectrum analysis provides a new method for the diagnosis of cancer.Serum(plasma) tumor markers play a more and more important role in the cancer census,diagnosis, prognosis,turnover judgment,therapeutic effect evaluation and follow-up.The effect of tumor markers in tumor especially in lung cancer,has been generally accepted. However,either fluorescence spectrum or tumor markers cannot solve the problem because both of them contain multi-parameter which cannot be efficiently analyzed by general statistics analysis methods.Artificial neural network(ANN) is a newly emerging cross-edge science that is related to biology,electronics,computer,mathematics and physics.As a new information processing technology,ANN has evident superiority in solving nonlinear, multi-parameter and uncertain complicated problems.ANN is one of fuzzy information processing systems,it not only does not require the detailed mechanism, but also has good self-adaptive capacity,self-organization leaming ability and fault tolerance function.Through training for actual data record,ANN can simulate and prediction system behaviors fairly precisely.In recent years,pattern recognition, discrimination and prediction based on ANN have been widely used in biomedicine. Also ANN has been used in tumor clinical prognosis,computer aided diagnosis and survival analysis.Domestic and foreign researchers have implemented computer aided diagnosis in lung cancer through ANN model,and have obtained good effect.Our objective is to establish a lung cancer ANN model that makes use of ANN technology platform combined with fluorescence spectrum and tumor markers for diagnosis of lung cancer as a clinical assistant method.Materials and Methods1.129 serum and plasma samples(42 malignant,42 benign and 45 normal).2.Analysis software of Matlab7.0 and SPSS12.0.3.Detection of serum fluorescence spectrum.Serum fluorescence spectrum was scaned under the wavelength of 405nm of excitation light.Fluorescence relative intensity parameters were selected every 5nm wavelength from 450nm to 700nm in the fluorescence emission,then extracted the principal components by principal component analysis.4.Determination of serum or plasma tumor markers.CEA,NSE,SCC-Ag and CYFRA21-1 were detected by the kits respectively;p16 was detected by methylation specific PCR.5.Construction of ANN models.Use the fluorescence spectrum parameters,the serum(plasma) tumor markers and the combination of both as the input layer neurons,respectively.Back-propagation algorithm was applied to develop ANN models,and then the fluorescence spectrum ANN model,the tumor markers ANN model and the fluorescence spectrum combined with tumor markers ANN model were established. 6.Compare the performance of the fluorescence spectrum combined with tumor markers ANN models with Fisher linear discriminatory analysis by ROC.Results1.According to principle of principal component characteristic value,3 fluorescence spectrum principal components were extracted.2.The numbers of 3 ANN models input layer parameters are 8,3 and 5,and the numbers of the 3 ANN models imply neurons are 8,3 and 4,respectively;the 3 ANN models were trained until the expected goal has been achieved.The prediction consistency rate of the 3 ANN models for training set is all 100%.3.The sensitivity of the 3 ANN models for prediction are 86.7%,60.0%and 66.7%; the specificity is 96.8%,80.6%and 83.9%;the consistency rate are 89.1%,67.4% and 69.6%;the positive predictive value are 92.9%,60.0%and 62.5%;the negative predictive value are 93.8%,80.6%and 80.0%;the area under ROC are 0.972,0.759 and 0.852,respectively,(P＜0.05).4.The sensitivity of fluorescence spectrum combined with tumor markers ANN model and Fisher linear discriminatory analysis are 92.9%and 62.9%;the specificity are 98.9%and 92.0%;the consistency rate are 96.1%and 76.7%;the positive predictive value are 97.6%and 78.8%;the negative predictive value are 97.7%and 75.5%;the area under the ROC curve are 0.996 and 0.787,respectively, (P＜0.05).5.The sensitivity of the 2 ANN models are 92.9%and100%;the specificity are 98.9%and 98.5%;the consistency rate are 96.1%and 96.9%;the positive predictive value are 97.6%and 97.9%;the negative predictive value are 97.7% and 100%,respectively.Conclusion1.The prediction result of the fluorescence spectrum combined with tumor markers ANN model is superior to that of the fluorescence spectrum ANN model or the tumor markers ANN model.2.The performance of ANN model is superior to that of Fisher linear discriminatory analysis.3.ANN can be used in diagnosis of lung cancer as a clinical assistant method.