Imaging Study Of Molecular Typing Of Hepatocellular Carcinoma Using Serial Algorithm Of Image Segmentation Slices

Objective: CT images of patients with hepatocellular carcinoma(HCC)were used to construct a model by deep learning method,and the image segmentation serial algorithm was applied to predict the molecular classification of patients with HCC,and the superiority and feasibility of the method was determined by comparing with other models.Compared with the bioinformatics model,the accuracy of its prediction under non-invasive conditions was explored,aiming to better evaluate the prognosis of patients with hepatocellular carcinoma when their physical conditions do not tolerate surgical or nonsurgical biopsy,and patients refuse invasive treatment,determine scientific and reasonable diagnosis and treatment plans,timely follow-up,reduce the pain of patients and reduce the economic burden.It has certain reference value for assisting doctors to determine clinical diagnosis and treatment plan.Methods: CT image data of hepatocellular carcinoma patients were obtained from The cancer genome atlas(TCGA)and the classification information of TCGA hepatocellular carcinoma patients were matched.Co-Scale Conv-Attentional Image Transformers(CoaT)was used to model 62 hepatocellular carcinoma patients with both image data and molecular typing information.The results were analyzed and compared with Alex Net model to verify the high performance of CoaT model.At the same time,the gene expression data of 183 patients with subtype information was screened by HSIC Lasso for transcriptional features related to molecular typing,and the model was established and compared.Results: CoaT was used to model CT images of hepatocellular carcinoma to achieve the purpose of molecular classification.The classification effect was good,and the model accuracy reached 94.9%.Compared with Alex Net model,when the epoch is 300,CoaT’s precision is 20.7%,18.3% and 7.4% higher than Alex Net in the three categories,recall is7.4%,23.3% and 18.5% higher,F1 score is 0.145,0.21 and 0.131 higher,respectively.The accuracy is 15.9% higher.In contrast,CoaT classification model showed better accuracy,precision,recall and F1 score in 300 epoch.It can be seen that CoaT has a better effect than Alex Net.In order to compare the effect of CoaT model with the traditional model characterized by transcription features,the HSIC Lasso model was created.When the epoch of CoaT model is 200,the area under the ROC(Receiver Operating Characteristic)curve of class 1is 0.89.The area under ROC curve of class 1 in HSIC Lasso model was 0.869.Both CoaT and bioinformatics model have good effects.CoaT model has certain practical value in determining molecular typing of liver cancer images.Conclusion: CoaT image segmentation serial algorithm can be used for molecular classification of CT images of hepatocellular carcinoma with high accuracy.Deep learning method can be well applied to CT image data to achieve molecular classification detection of hepatocellular carcinoma,which can better assess the prognosis of patients,determine clinical diagnosis and treatment plans,and guide patients to adopt scientific and reasonable diagnostic methods and treatment plans.