Machine learning-based prognostic prediction for cutaneous malignant melanoma patient survival.

BACKGROUND

Cutaneous malignant melanoma (CMM) is a highly malignant tumor that necessitates early diagnosis and precise survival prediction. The development of accurate prognostic models is essential for improving patient survival.

RESEARCH DESIGN AND

METHODS

This retrospective study analyzed data from 5979 CMM patients in the SEER database (2004-2015), with external validation using the TCGA dataset. Patients were randomly allocated to training and testing sets in a 7:3 ratio. The SMOTE+DeepSurv (DeepSmote) model was compared against seven models, including DeepSurv, XGBoost, Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Decision Tree (DT). Model performance was evaluated using Area Under the Curve (AUC), accuracy, precision, recall, and F1-score.

RESULTS

The DeepSmote model demonstrated superior prognostic performance across both SEER and TCGA datasets. On the SEER test set, it achieved an AUC of 0.96, accuracy of 0.95, and F1-score of 0.95 for 1-year prediction. This strong performance was maintained in the external TCGA cohort (AUC: 0.91, accuracy: 0.88, F1-score: 0.87), and consistent superiority was observed for 3- and 5-year predictions, confirming its robustness and generalizability.

CONCLUSION

DeepSmote provides an accurate, generalizable prognostic tool for CMM survival prediction, outperforming other models across multiple datasets and evaluation metrics.