Establishing a Differential Diagnosis Method for Melanoma Using Blood Parameters Based on Machine Learning Technology

Objective: This study aims to develop and validate a differential diagnosis model for melanoma. This model will combine blood test results with machine learning techniques to provide an inexpensive, non-invasive method for early melanoma screening. Methods: Complete blood counts and biochemical indicators were retrospectively collectedfrom 4,534 patients for clinical data. Key variables were screened using the Recursive Feature Elimination (RFE) and Mutual Information (MI) methods. A predictive model was established using the XGBoost algorithm, and the model's performance was evaluated through 10-fold cross-validation and an independent test set. Results: The model achieved an area under the curve (AUC) of 0.996 on the training set and 0.975 on the test set, with respective accuracies of 0.981 and 0.926. MPV, DBIL, CK, Na, HDL-C, and LDH were identified as the most predictive indicators. Conclusion: This study confirms that conventional blood parameters combined with machinelearning technology can effectively distinguish melanoma, demonstrating high diagnostic value. This approach provides a feasible, noninvasive strategy for earlymelanoma identification.