Translational cancer research · 2026
Development and internal validation of machine learning models for predicting lymph node metastasis in cutaneous head and neck melanoma.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Cheng Bingyu, Yu Jianqun, Ji Qiang
Аннотация
BACKGROUND: Cutaneous head and neck melanoma (CHNM) carry substantial metastatic potential, yet reliable individualized prediction of lymph node metastasis (LNM) remains challenging for clinical practice. This study developed machine learning models to predict LNM risk in patients with CHNM using routine clinicopathological data, and compared model predictive performance to support clinical nodal risk stratification for sentinel lymph node biopsy decision-making.
METHODS: We enrolled 1,451 patients with CHNM. Among them, 242 (16.7%) had LNM, and 1,209 (83.3%) did not. We used one-hot encoding for categorical variables and Z-score standardization for continuous variables, respectively. The Boruta algorithm was used to select key factors. Seven machine learning models were built, with hyperparameters optimized by 5-fold cross-validation. We tested model performance using sensitivity, specificity, area under the curve (AUC), calibration curves, and decision curve analysis (DCA) eta. Finally, SHapley Additive exPlanations (SHAP) analysis was adopted to explain model results.
RESULTS: According to the Boruta algorithm, age, histological type, ulceration, mitotic rate and Breslow thickness were selected as independent predictive factors. Nodular melanoma (Histologic_X8721) emerged as a novel independent predictor, which has been rarely reported in CHNM. The extreme gradient boosting (XGBoost) showed good AUC of 0.748 [95% confidence interval (CI): 0.692-0.804], balanced performance (accuracy 0.686, sensitivity 0.658, specificity 0.691), and better clinical practicability in high-threshold DCA. KNN achieved the highest sensitivity (0.671), alongside a balanced accuracy of 0.687 and an AUC of 0.734 (95% CI: 0.672-0.795), with good calibration and net benefit in most situations. Although the MLP obtained the highest AUC value (0.755, 95% CI: 0.698-0.811), the model showed sub-optimal outcomes on calibration and decision-curve-analysis plots.
CONCLUSIONS: These interpretable machine learning models may serve as an auxiliary reference for assessing LNM risk at initial diagnosis, supporting nodal metastasis risk stratification and clinical judgment of sentinel lymph node biopsy in CHNM patients.
METHODS: We enrolled 1,451 patients with CHNM. Among them, 242 (16.7%) had LNM, and 1,209 (83.3%) did not. We used one-hot encoding for categorical variables and Z-score standardization for continuous variables, respectively. The Boruta algorithm was used to select key factors. Seven machine learning models were built, with hyperparameters optimized by 5-fold cross-validation. We tested model performance using sensitivity, specificity, area under the curve (AUC), calibration curves, and decision curve analysis (DCA) eta. Finally, SHapley Additive exPlanations (SHAP) analysis was adopted to explain model results.
RESULTS: According to the Boruta algorithm, age, histological type, ulceration, mitotic rate and Breslow thickness were selected as independent predictive factors. Nodular melanoma (Histologic_X8721) emerged as a novel independent predictor, which has been rarely reported in CHNM. The extreme gradient boosting (XGBoost) showed good AUC of 0.748 [95% confidence interval (CI): 0.692-0.804], balanced performance (accuracy 0.686, sensitivity 0.658, specificity 0.691), and better clinical practicability in high-threshold DCA. KNN achieved the highest sensitivity (0.671), alongside a balanced accuracy of 0.687 and an AUC of 0.734 (95% CI: 0.672-0.795), with good calibration and net benefit in most situations. Although the MLP obtained the highest AUC value (0.755, 95% CI: 0.698-0.811), the model showed sub-optimal outcomes on calibration and decision-curve-analysis plots.
CONCLUSIONS: These interpretable machine learning models may serve as an auxiliary reference for assessing LNM risk at initial diagnosis, supporting nodal metastasis risk stratification and clinical judgment of sentinel lymph node biopsy in CHNM patients.