Machine Learning Approaches for Predicting Radiation Therapy Outcomes: A Clinician's Perspective

Area of Science:

• Radiation oncology
• Medical modeling
• Machine learning

Background:

• Radiation oncology has historically relied on modeling, evolving with data and genomics.
• Machine learning (ML) methods, widely used in tech, are increasingly applied in medicine for prediction.
• Clinical adoption of ML in radiation oncology is hindered by the complexity of these models for clinicians.

Purpose of the Study:

• To provide a clinician-friendly review of ML for predicting radiation therapy outcomes.
• To lower the barrier to entry for clinicians without formal ML training.
• To discuss principles for evaluating and creating ML models in radiation oncology.

Main Methods:

• Review of ML applications in predicting radiation therapy outcomes.
• Description of 7 key principles for evaluating/creating ML models.
• Introduction and critique of 3 popular ML methods: logistic regression (LR), support vector machine (SVM), and artificial neural network (ANN).

Main Results:

• Multiple research groups have demonstrated the value of applied ML in radiation oncology.
• Current studies are in exploratory phases, but methodology is maturing.
• The field is progressing towards readiness for larger-scale investigations.

Conclusions:

• ML holds significant promise for improving prognostic and therapeutic applications in radiation oncology.
• Understanding ML principles and common methods can facilitate clinical adoption.
• Further large-scale research is warranted to fully leverage ML in radiation therapy.