A Glycomics-based Liver Cancer Diagnosis Model Based On Self-encoder And Machine Learning

Primary hepatocellular carcinoma is one of the most common malignant tumors worldwide,second only to lung cancer in China,and the incidence rate is increasing year by year,and now it has become one of the important causes of death from malignant tumors in China.Early detection plays an important role in the diagnosis and later treatment of liver cancer,and if accurate prediction of early liver cancer can be made,it can help control liver cancer.Previously,early detection of liver cancer mainly relied on blood tests and imaging techniques,but recent advances in computer-aided technology have promoted the application of machine learning methods in the field of cancer diagnosis.In this thesis,a combined model based on self-encoder and machine learning was used to collect blood glycomics data from1618 liver disease patients in a tertiary general hospital in Chongqing using fluorescent capillary oligosaccharide chain detection,and the relatively optimal model was selected by five model evaluation indexes: accuracy,precision,recall,F1 value and AUC value,finally establishing a model that can diagnose early stage This model can be used to diagnose early stage liver cancer,and provide an auxiliary diagnosis for clinical liver cancer early diagnosis.Compared with the existing liver cancer prediction models,this thesis tries to improve the accuracy of the model from three aspects.First,the number of liver cancer patient samples under normal conditions is much smaller than that of non-hepatocellular carcinoma patients,and the SMOTE method is used to generate liver cancer samples to solve the problem of data imbalance.Second,the original data is reconstructed to some extent by self-encoder to extract the implied information features and obtain the hidden layer data features.Third,the self-encoder and machine learning algorithm are combined,and the reconstructed data are applied to the machine learning algorithm,so that the performance of the machine learning algorithm can be optimized to some extent.By comparing the evaluation indexes of the single machine learning model and the combined self-encoder-machine learning model,it is found that the prediction effect of the combined model is significantly better than that of the single model,among which the combined self-encoder-random forest model has the best judgment result.