Knowledge-light adaptation approaches in case-based reasoning for radiotherapy treatment planning

Area of Science:

• Medical Physics
• Artificial Intelligence
• Oncology

Background:

• Radiotherapy treatment planning is a complex, time-consuming process requiring expert collaboration.
• Current case-based reasoning (CBR) systems lack adaptation, necessitating domain-specific knowledge for new patient cases.
• Knowledge-light adaptation approaches are explored to overcome the limitations of acquiring extensive domain knowledge.

Purpose of the Study:

• To enhance a previously developed case-based reasoning (CBR) system for brain tumor radiotherapy treatment planning.
• To investigate knowledge-light adaptation methods that minimize the need for domain-specific expertise.
• To improve the efficiency and accuracy of radiotherapy treatment plan generation.

Main Methods:

• Developed two knowledge-light adaptation approaches: machine learning (neural networks, naive Bayes) and adaptation-guided retrieval.
• Applied these methods to adapt beam number and beam angles in retrieved radiotherapy treatment plans.
• Evaluated adaptation strategies using real-world brain cancer patient cases treated with 3D-conformal radiotherapy.

Main Results:

• Neural network-based adaptation improved the CBR system's success rate by 12% compared to no adaptation.
• Adaptation-guided retrieval for beam number increased the success rate by 29%.
• Naive Bayes classifier did not improve results; adaptation-guided retrieval showed limited success for beam angle adaptation.

Conclusions:

• The proposed adaptation methods effectively improve CBR system performance for recommending the number of beams.
• Successful adaptation of beam angles requires a substantial and relevant case base.
• Knowledge-light adaptation shows promise for optimizing radiotherapy treatment planning.