Premium
SU‐E‐T‐48: Application of Evidence Theory in Radiation Oncology Outcome Analysis
Author(s) -
Chen W,
Cui Y,
Galvin J,
Yu Y,
Hussaini Y,
Xiao Y
Publication year - 2011
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.3611999
Subject(s) - outcome (game theory) , range (aeronautics) , statistics , mathematics , statistical theory , bounded function , computer science , medical physics , econometrics , medicine , mathematical economics , materials science , composite material , mathematical analysis
Purpose: To apply evidence theory to radiotherapy outcome analysis for better combination of clinical evidence, some of them not self‐consistent, incorporating uncertainties of the evidence. To illustrate the theory, we used clinical data of different sources from Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC) publications. The belief‐plausibility range obtained from the combined outcome presents a fundamentally new method for clinical outcome evaluation. Methods: Evidence theory was applied to the original data obtained from QUANTEC. In evidence theory the uncertainty of practical data is characterized by the basic belief assignment function, therefore the uncertainty is quantitatively included in the statistical calculations. Considering the inconsistent and conflicting data from different sources, we used Yagerˈs combination rule to combine the statistical information and obtained a belief‐plausibility range for the dose‐volume outcome. The goodness of the predictive power is evaluated with receiver operating characteristic (ROC) graph. Results: With evidence theory and Yagerˈs combination rules, we fused the data from different sources with respective uncertainties and obtained the belief‐plausibility range, which can be considered as a probability range bounded by its lower and upper limits. For the example of predicting pneumonitis from lung cancer treatments (data from three institutions are combined, cf. supporting document). Logistic fits for the values of belief and plausibility functions at different doses were obtained. Two commonly used NTCP models, LKB and CV model, are applied and fitted to clinical data of pneumonitis respectively. A probability range of uncertainties is obtained for each model. Evidence theory is used to fuse the strength of these two models. Conclusions: We have presented two examples to demonstrate how evidence theory can be used to combine clinical evidence of various uncertainties. With evidence theory, belief and plausibility are utilized for clinical evidence evaluation, a new alternative to the conventional probability presentation of the clinical outcome.