MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Forecasting Individual Patient Response to Radiotherapy with a Dynamic Carrying Capacity Model

M Zahid*1, N Mohsin1, A Mohamed2, J Caudell1, L Harrison1, C Fuller2, E Moros1, H Enderling1, (1) H. Lee Moffitt Cancer Center, Tampa, FL, (2) UT M.D. Anderson Cancer Center, Houston, TX

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: To simulate tumor growth and response to radiotherapy by modeling the effect of radiotherapy (RT) as a reduction in the tumor carrying capacity, indicative of the impact of RT on the tumor microenvironment, and to make predictions of individual patient responses to RT and related clinical outcomes.

Methods: Tumor volume data were collected for 2 independent cohorts (training and cross-validation) of head and neck cancer patients that received 66-70 Gy in 2 Gy daily fractions. Clinical outcome data, i.e. locoregional control (LRC) and disease-free survival (DFS), were also collected. Tumor growth was modeled as logistic growth with one parameter (?, volumetric growth rate), and the effect of each RT dose was modeled as an instantaneous reduction in the carrying capacity with one parameter (d, carrying capacity reduction fraction).

Results: The model fit data from the training cohort with two patient-specific parameter values with high accuracy (R² = 0.95). Model analysis revealed that growth rate is not patient specific. A uniform ? reduced to R² = 0.92 while reducing the number of free parameters in the model to one. This model fit the cross-validation data with high accuracy (R² = 0.98), demonstrating transferability of ?. To predict response to RT, we combined the d-distribution from the training cohort and measurements of volume reduction in individual patients from the cross-validation cohort to estimate d. This method predicts patient-specific outcomes in the cross-validation cohort with >95% accuracy for LRC and >87% accuracy for DFS with the inclusion of 2 weekly volume measurements.

Conclusion: This work shows that the impact of RT on the tumor microenvironment may be sufficient to model patient-specific tumor volume dynamics. Additionally, the prediction pipeline shows a remarkable capacity to predict clinical outcomes with high accuracy in an independent cohort with the inclusion of just a few patient measurements.

Keywords

Cross Validation, Modeling, ROC Analysis

Taxonomy

TH- Dataset Analysis/Biomathematics: Informatics

Contact Email