A pilot study of machine-learning based automated planning for primary brain tumours

Area of Science:

• Radiation Oncology
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• High-quality radiotherapy (RT) planning is crucial for minimizing late effects in pediatric and young adult brain tumor patients.
• Automated machine learning (ML) feasibility for RT planning in this cohort was previously unstudied.

Purpose of the Study:

• To assess the feasibility of using an automated ML model to aid in RT planning for primary brain tumors in children and young adults.
• To compare ML-generated RT plans with clinically delivered (human-generated) plans.

Main Methods:

• A ML model was trained on 95 patients' data to predict dose distributions based on image features.
• The ML model was tested on 15 previously treated patients to generate predicted dose distributions.
• Dosimetry for target volumes and organs-at-risk (OARs) were compared between ML and manual plans.

Main Results:

• The ML method successfully generated deliverable RT plans for all 15 test patients within 30 minutes.
• All ML plans achieved 95% of the prescription dose to the planning target volume.
• OAR doses were comparable, with ML plans showing lower mean doses to the brain and left temporal lobe but higher doses to the right cochlea and lenses.

Conclusions:

• Automated ML is a feasible tool for aiding RT planning in pediatric and young adult brain tumor patients.
• ML-generated plans achieved high quality, delivering the prescribed 54 Gy dose effectively.
• Further clinical implementation and evaluation of this ML approach are warranted.