基于双尺度深度学习的肝细胞癌常规病理图像基因突变预测OA

Objective To explore the clinical value of utilizing deep learning and routine whole slide images(WSIs)for the non-invasive prediction of key gene mutations in hepatocellular carcinoma(HCC).Methods A retrospective analysis was conducted on 217 HCC WSIs from The Cancer Genome Atlas(TCGA)database Liver Hepatocellular Carcinoma(LIHC)cohort database,which were randomly divided into a training set(n=174)and an internal testing set(n=43).Furthermore,100 HCC patients from the First Affiliated Hospital of Wenzhou Medical University were included as an independent external validation set.Following image standardization pretreatment such as color normalization,a"macro-micro"dual-scale feature fusion deep learning model was constructed.The macroscopic branch of this model extracted tissue texture features via attention-based multiple instance learning(MIL),while the microscopic branch extracted cellular spatial topological features through nuclear segmentation combined with a graph convolutional network(GCN).The fused dual-scale features were then applied to the end-to-end prediction of the mutation status of either the TP53 or CTNNB1 key gene.Results In the internal testing set,the accuracy of the dual-scale fusion model for predicting the mutation of either gene reached 81.9%,with an area under the curve(AUC)of 0.848,a sensitivity of 80.0%,and a specificity of 84.6%.Ablation studies confirmed that the introduction of microscopic features significantly enhanced the model's capacity to capture tumor heterogeneity(AUC increased from 0.641 to 0.848,P<0.05).In the cross-center external validation set,the model achieved a predictive AUC of 0.812 and an accuracy of 78.5%.Conclusion The proposed dual-scale deep learning model demonstrates potential in non-invasively predicting key HCC gene mutations solely based on routine pathological slides.It holds promise as a cost-effective,preliminary screening tool for individualized clinical diagnosis and treatment.Although its precise clinical efficacy warrants further verification through prospective large-sample studies.