Artificial Intelligence for Dose-Volume Histogram Based Clinical Decision-Making Support System in Radiation Therapy Plans for Brain Tumors
P Siciarz*, B McCurdy, S Alfaifi, S Rathod, E Van Uytven, R Koul, CancerCare Manitoba, Winnipeg, MB, Canada
Presentations
WE-D-TRACK 1-5 (Wednesday, 7/28/2021) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]
Purpose: To create and investigate a novel, clinical decision-support system using artificial intelligence (A.I.).
Methods: The A.I. model was developed based on 79 radiotherapy plans of brain tumor patients that were prescribed a total dose of 60 Gy delivered in 30 fractions (2 Gy per fraction) with volumetric-modulated arc therapy (VMAT). Structures considered for analysis included planning target volume (PTV), brainstem, cochleae, and optic chiasm. Data for model training consisted of five dosimetric and five geometric features extracted from radiation therapy plans. The model aimed to classify the target variable that included class-0 corresponding to plans for which the PTV treatment planning objective was met and class-1 that was associated with plans for which the PTV objective was not met due to the priority trade-off to meet one or more organs-at-risk constraints. One model was selected for further investigation amongst those evaluated: Support-Vector Machine, Elastic-Net, Logistic-Regression, and Random-Forest using double-nested cross-validation and an area under the curve (AUC) metric. The model predictions were explained both globally and locally with Shapely additive explanation (SHAP) interaction values.
Results: The highest-performing model was Logistic Regression achieving an accuracy of 93.8±4.1% and AUC of 0.98±0.02 on the testing data. The SHAP analysis indicated that the ΔD99% metric for PTV had the greatest influence on the model predictions and was approximately 3.5 times more significant than the second most important feature – ΔD1% for optic chiasm. The least important feature was ΔDMAX for the left and right cochleae.
Conclusion: The trained model achieved satisfactory accuracy and can be used by medical physicists in a data-driven quality assurance program as well as by radiation oncologists to support their decision-making process in terms of treatment plan approval and potential plan modifications. Model explanation analysis showed that the model relies on clinically valid logic when making predictions.
Handouts
Keywords
Dose Volume Histograms, Radiation Therapy
Taxonomy
Contact Email