Fig 1.
*According to the 3rd edition of the International Classification of Tumors code (Site recode ICD-O-3/WHO 2008). **8721/3 represents nodular melanoma. ML,Machine Learning; MLP, Multilayer Perceptron.
Table 1.
Patient baseline information.
Table 2.
Univariate and multivariate logistic regression.
Fig 2.
(A) Ten-fold cross-validation of the six machine learning models. (B) The ROC curves of the six machine learning models. 1-Specificity indicates the false positive rate of the model and Sensitivity indicates the true positive rate. (C) The ROC curves of the MLP. (D) The radar plots of the six machine learning models. AUC, the area under the curve; ROC, Receiver Operator Characteristic; MLP, Multilayer Perceptron; AB, Adaptive Boosting; BAG, Bagging; LR, Logistic regression; GBM, Gradient Boosting Machine; XGB, eXtreme Gradient Boosting.
Fig 3.
The confusion matrix for six algorithmic models.
MLP, Multilayer Perceptron; AB, Adaptive Boosting; BAG, Bagging; LR, Logistic regression; GBM, Gradient Boosting Machine; XGB, eXtreme Gradient Boosting.
Table 3.
Comparison of six machine learning methods.
Fig 4.
Explanation of feature importance for MLP model using SHAP.
(A) SHAP summary plot. X-axis is determined by each Shapley value. (B)(C) SHAP force plot. Features predicted to increase the output are depicted in red, while those predicted to decrease the output are in blue. The length of an arrow correlates directly with the magnitude of the feature’s impact on the output. MLP, Multilayer Perceptron; SHAP, SHapley Additive exPlanation.
Table 4.
Univariate and multivariate Cox regression.
Fig 5.
Forest plots for univariate and multivariate Cox regression.
Fig 6.
Nomogram of 1-year, 3-year, and 5-year OS for patients with nodular melanoma.
Prognostic variables for a specific patient are aligned with the top "Points", generating individual scores for each variables. Subsequently, the sum of these scores is aligned with the bottom "Total Points" to estimate the patient’s overall survival probability. OS, overall survival; MLP, multilayer perceptron.
Fig 7.
(A) (B) (C) OS calibration curves at 1, 3 and 5 years. (D) DCA curves for prognostic models. (E) ROC curves for 1-, 3- and 5- year OS. DCA, Decision Curve Analysis; ROC, Receiver Operator Characteristic.
Fig 8.
Kaplan-Meier curves for each prognostic factor in patients with nodular melanoma.
Fig 9.
(A) Risk score grouping. (B) Risk scores and patient survival time. (C) Risk heat map. The green line represents the low-risk stratum and the red line represents the high-risk stratum.