A Guide to Cross-Validation for Artificial Intelligence in Medical Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Artificial intelligence (AI) is increasingly integral to medical imaging technologies.
• Accurate AI algorithm development necessitates robust prediction error estimation.
• Cross-validation (CV) is a key technique to mitigate bias and overoptimism in AI models.

Purpose of the Study:

• To introduce the principles of cross-validation (CV) for AI in medical imaging.
• To provide a practical guide for applying CV in AI algorithm development.
• To discuss various CV techniques, their pros and cons, and common pitfalls.

Main Methods:

• Explanation of cross-validation (CV) principles.
• Description of different CV techniques (e.g., k-fold, leave-one-out).
• Guidance on practical application and avoiding common errors in prediction error estimation.

Main Results:

• CV is crucial for preventing overoptimism in AI algorithms.
• CV helps mitigate biases in hyperparameter tuning and algorithm selection.
• Understanding different CV methods allows for appropriate technique selection based on the scenario.

Conclusions:

• Cross-validation (CV) is a fundamental tool for reliable AI development in medical imaging.
• Proper application of CV ensures more accurate prediction error estimation.
• This guide aids researchers in effectively using CV for supervised learning and CNN development.