Incorporating Case-Based Reasoning for Radiation Therapy Knowledge Modeling: A Pelvic Case Study.

Area of Science:

• Medical Physics
• Radiotherapy
• Machine Learning

Background:

• Existing radiotherapy knowledge models struggle with evolving treatment schemes and novel patient anatomies.
• Accurate dosimetry prediction is crucial for effective radiotherapy treatment planning.

Purpose of the Study:

• To propose and evaluate a case-based reasoning (CBR) framework for improved radiotherapy dosimetry prediction.
• To handle novel patient anatomies beyond original training data in intensity-modulated radiotherapy (IMRT).

Main Methods:

• Developed a CBR framework to manage variations in patient anatomy.
• Analyzed 105 pelvic IMRT cases (80 prostate, 25 prostate-plus-lymph-node).
• Simulated four scenarios (Scarce, Semiscarce, Semiample, Ample) and compared CBR with stepwise regression using sum of squared residuals.

Main Results:

• CBR showed significantly lower mean sum of squared residuals for bladder predictions in outliers (0.174 vs. 0.459, P=.0326).
• For the rectum, CBR also demonstrated lower residuals (0.103 vs. 0.150, P=.1972).
• The Ample scenario, retaining novel cases, showed significant improvement for the bladder model over the Scarce scenario (P=.0398).

Conclusions:

• Case-based reasoning offers improved prediction accuracy and robustness for radiotherapy dosimetry.
• The proposed CBR framework effectively handles novel anatomies, enhancing treatment planning guidance.
• Integrating CBR with predictive models can optimize clinical radiotherapy practices.