Machine learning to predict overall short-term mortality in cutaneous melanoma.

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Cutaneous malignant melanoma (CMM) is a frequent malignancy where accurate staging is crucial for prognosis.
• Artificial intelligence (AI) is increasingly utilized for developing reliable prognostic staging systems for CMM.
• This study aimed to create a machine learning-based tool for predicting short-term survival in CMM patients.

Purpose of the Study:

• To develop and validate an AI-driven tool for predicting 3-year mortality in cutaneous malignant melanoma (CMM).
• To identify key clinicopathological variables influencing CMM survival beyond current staging systems.
• To deploy the best-performing AI model as an accessible online tool for clinical use.

Main Methods:

• Utilized CMM data from the Veneto Cancer Registry and regional health service.
• Employed univariate Cox regression to assess variable prognostic strength.
• Trained and evaluated multiple machine learning models, including Deep Neural Networks and Random Forests, using cross-validation and hyperparameter optimization.
• Assessed model performance using balanced accuracy, precision, recall, and F1 score on a separate test set.

Main Results:

• Univariate analysis confirmed the prognostic value of TNM staging and identified additional significant variables (sex, tumor site, histotype, growth phase, age).
• The Neural Network and Random Forest models demonstrated superior prognostic performance, achieving balanced accuracies of 91% and 88%, respectively.
• Key predictors of survival included age, T and M stages, mitotic count, and ulceration, as indicated by Gini importance scores.

Conclusions:

• An AI algorithm with high staging reliability for CMM was developed using routinely collected clinicopathological data.
• A web-based tool implementing this AI algorithm is available (unipd.link/melanomaprediction).
• The tool's minimal implementation requirements facilitate its testing and validation in clinical practice for enhanced patient management.