Using Machine Learning and Deep Learning Methods to Predict the Complexity of Breast Cancer Cases

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Multidisciplinary tumor boards (MTBs) are crucial for collaborative cancer patient management.
• Increasing case numbers challenge the efficiency of MTBs due to time limitations.
• Clinical decision support systems (CDSS) sometimes fail to provide recommendations for complex cases.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the complexity of breast cancer clinical cases.
• To identify complex cases that require in-depth discussion within MTBs.
• To enhance the efficiency of cancer care management by prioritizing complex cases.

Main Methods:

• Trained and tested various machine learning and deep learning algorithms.
• Utilized a labeled dataset of 298 breast cancer patient summaries.
• Evaluated model performance using metrics such as the F1 score.

Main Results:

• XGBoost and multi-layer perceptron models demonstrated the highest performance.
• Achieved an F1 score of 83% in predicting breast cancer case complexity.
• Identified specific machine learning models capable of distinguishing complex cases.

Conclusions:

• Machine learning models can effectively predict breast cancer case complexity.
• This approach can assist MTBs in managing time constraints and focusing on complex cases.
• The developed models offer a potential solution to improve the efficiency of cancer care discussions.