Prediction of Metastasis in Paragangliomas and Pheochromocytomas Using Machine Learning Models: Explainability...

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Paragangliomas and pheochromocytomas present a significant challenge due to a high rate of metastatic disease (up to 20%) with unreliable prediction.
• The 'black box' nature of many machine learning models hinders trust and clinical adoption, despite their potential for improved predictive accuracy.

Purpose of the Study:

• To develop and validate a machine learning architecture integrating data mining and explainability techniques for predicting metastatic disease in paragangliomas and pheochromocytomas.
• To enhance trust in predictive models by understanding the factors driving their predictions, facilitating clinical integration.

Main Methods:

• A comprehensive data preprocessing pipeline was implemented.
• Various classification algorithms were applied, with a focus on interpreting their outputs.
• Explainability techniques and feature selection were integrated to identify key predictive variables and assess their impact on model performance.

Main Results:

• The Random Forest algorithm demonstrated superior performance, achieving 96.3% accuracy, 96.5% precision, and an AUC of 0.963.
• The study successfully identified key variables influencing metastatic disease prediction.
• Integrated explainability provided insights into the model's decision-making process.

Conclusions:

• The proposed machine learning architecture effectively predicts metastatic disease in paragangliomas and pheochromocytomas with high accuracy.
• Combining predictive performance with model explainability fosters trust and supports the integration of these tools into clinical practice.
• This approach offers a promising avenue for improving patient management and outcomes for these rare tumors.