Developing an ensemble machine learning study: Insights from a multi-center proof-of-concept study.

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning for Medical Imaging
• Radiomics and Cancer Diagnostics

Background:

• Machine learning models in medical imaging show promise but often function as isolated tools.
• Existing algorithms for similar diagnostic tasks could be integrated to improve performance.
• An ensemble approach offers a method to aggregate diverse algorithms for enhanced classification.

Purpose of the Study:

• To develop and validate an ensemble approach for integrating multiple machine learning algorithms.
• To improve classification performance in discriminating metastatic from non-metastatic lung cancer patients.
• To provide an interpretable framework for ensemble model predictions.

Main Methods:

• Utilized a public database of radiomic features from CT scans of 535 lung cancer patients.
• Trained seven independent machine learning algorithms to classify metastatic versus non-metastatic patients.
• Integrated algorithm outputs using a Support Vector Machine (SVM) classifier and applied Explainable Artificial Intelligence (XAI).

Main Results:

• The ensemble model achieved higher accuracy compared to individual algorithms, with an accuracy of 0.78 on an independent test set.
• The ensemble model yielded a F1-score of 0.57 and a log-loss of 0.49.
• Shapley values provided insights into individual algorithm contributions and methodological impacts, enhancing model interpretability.

Conclusions:

• The proposed ensemble approach offers an innovative method for integrating existing algorithms.
• This framework lays the groundwork for future evaluations in diverse clinical scenarios.
• The ensemble model enhances diagnostic accuracy and interpretability in lung cancer classification.