Fig. 8From: Early detection of hepatocellular carcinoma via no end-repair enzymatic methylation sequencing of cell-free DNA and pre-trained neural networkDeepTrace outperformed other models in detection of HCC individuals in the validation cohort. A ROC curve of different models on HCC detection (n = 130). Blue lines show the average ROC of four-fold cross-validation. B–D Accuracy, sensitivity, and specificity of different models on HCC detection (62 HCC, 48 LD, 20 healthy). The boxplots show the data distribution of four-fold cross-validation results. E,F Results of DeepTrace and AFP biomarker on HCC detection with BCLC staging system (E) and TNM staging system (F) (61 HCC, 48 LD). An HCC patient without AFP value was excluded from the analyses in panels E and F. The error bars indicate the standard deviation of four-fold cross-validation results. The numbers near to the boxplots and histograms in panels B–F are the mean valuesBack to article page