Algorithm for Patient-specific Treatment Plan

Patient-Specific Replan Probability Prediction for Radiotherapy Treatment

Application

A replan-probability-constrained optimization algorithm for patient-specific treatment plan optimization to reduce replan rates in radiation therapy.

Key Benefits

Ability to predict if a head and neck cancer patient will need a re-plan.
Determines the statistical significance of dosimetry and clinical features.

Market Summary

Intensity-modulated Proton therapy (IMPT) is an essential tool for treating malignancies, like head and neck, brain & skull-based cancers, and tumors near radiation-sensitive tissues such as the nerves and the gastrointestinal tract. IMPT enables precise radiation delivery to tumors while sparing healthy organs and tissues. Although the technology provides dosimetric advantages, it is highly sensitive to uncertainties in particle range and anatomical changes, especially from interfractional variations in patient positioning, organ motion, or target volume changes throughout the treatment. These limitations can prevent complete tumor control and lead to radiation exposure to healthy tissues.

Technical Summary

The invention consists of a replan-probability-constrained optimization algorithm utilizing neural networks (NN), radiomics, and dosiomics trained to predict if a head and neck (H&N) patient (or other treatment sites) would require a replan. The NN, radiomics, and dosiomics models could be implemented using computer programming languages, like Python, C#, C, etc. The input for all the models includes a) the patients’ medical and health characteristics, b) anatomical information in planning CT, quality assurance CT, and cone-beam CT images, and c) isodose distribution and dose volume histograms in the treatment plans. In addition, the contours of the targets and organ at risks (OARs), their corresponding beam specific water equivalent thickness (WET) and their changes will also be included for model training the prediction. The predictors for replans we found include the average beam heterogeneity, average max dose change, and time between surgery and pCT, target WET changes. Significant categorical features were tumor site, chemotherapy status, and surgery status.