Machine Learning Principles for Radiology Investigators.

Area of Science:

• Radiology and Medical Imaging
• Computer Science and Artificial Intelligence

Background:

• Artificial intelligence (AI) and deep learning (DL) are currently prominent in radiology research.
• Machine learning (ML) offers a broader range of statistical techniques that can complement DL approaches.

Purpose of the Study:

• To provide a refresher on basic statistical concepts relevant to ML.
• To review key ML considerations including regression, classification, decision boundaries, and the bias-variance tradeoff.
• To introduce advanced ML techniques and their applications in radiology.

Main Methods:

• Review of fundamental statistical concepts.
• Discussion of practical ML considerations: regularization, ground truth, populations, compute, and data management.
• Overview of advanced ML algorithms: bootstrapping, bagging, boosting, decision trees, random forest, XGboost, and support vector machines.

Main Results:

• The study outlines essential statistical concepts and ML principles.
• It details advanced ML techniques with potential applications in radiological studies.
• Examples from the radiology literature are provided to illustrate the use of these methods.

Conclusions:

• Machine learning provides a diverse set of statistical tools valuable for radiology research.
• Understanding these techniques can enhance the capabilities of investigators beyond deep learning.
• The review aims to equip researchers with knowledge of ML methods for improved radiological data analysis.