Background

Dosimetry of patients undergoing diagnostic radiology is vital for patient dose optimisation. It is therefore important that uncertainties in the determination of organ doses are limited or accurately determined for guidance. Therefore, this study estimated the uncertainties in estimated organ doses for computed tomography (CT) examinations.

Methods

The Bayesian hierarchical model built on PyMC probabilistic programming version 4.1.4, incorporated prior information (i.e. proportion and fraction irradiation) from scanning the anthropomorphic Alderson female rando phantom by using whole organ exposure modeling approach aiming to enhance the accuracy of dose estimation for selected organs in the head, chest and abdomen regions. Thermoluminescence dosimeter (TLD) measurements of organ doses of the anthropomorphic Alderson female rando phantom was integrated with posterior distribution modeling results obtained from BAyesian Model Building Interface (BAMBI) version 0.9.1. The simulation of the models assumed two (2) scenarios i.e. exposure of organs (eye, brain, thyroid, breast, colon, kidney, stomach) to continuous beam of X-ray assigned a binary value of 1 or 0 for being exposed or not respectively and 90% absorption of radiation dose by majority of the organs in the scanned region. Arviz was then used for exploratory analysis of the results from the simulation.

Results

Convergence was observed for all the model parameters (i.e. standard deviation; fraction irradiated; proportion irradiated for colon, eye, kidney, salivary glands and stomach respectively) indicating that each Markov chain for the parameters had converged to the target probability distribution and had low autocorrelation. The whole organ exposure model showed a substantial effect of irradiation on specific organs, particularly the eye and thyroid. The proportion irradiated had a mean of 0.115 mGy and standard deviation of 0.046 indicated a moderate effect and a highest density interval of 3%-hdi_97% (0.032, 0.198), suggesting a more substantial effect than fraction irradiated with a mean of 0.047 mGy and standard deviation of 0.025 and a highest density interval of 3%-hdi_97% (0.007, 0.094)..The credible intervals for some organ-specific effects ( Liver, Kidney, Stomach) were wide with values of (0.323, 0.182,0.454) indicating more uncertainty in the organ dose estimates. High effective sample sizes (ESS) (12,639.0,13,411.0, 14,251.0) were within the acceptable range for ESS which is typically considered to be greater than 100 and r_hat values close to 1 indicated convergence, while values that are much greater than 1 indicated insufficient mixing between Markov chain. These values across all model parameters indicated that the model converged well, providing reliable estimates.

Conclusions

The model successfully estimated uncertainties in observable data (measured TLD dose) of organ doses from CT exposure. The hierarchical whole organ exposure modeling approach presented in this study demonstrated better results for improving the accuracy of organ dose estimation and showed a substantial effect of irradiation on specific organs, particularly the eye and thyroid.

Significance

The study:

Demonstrates better results for improving the accuracy of organ dose estimation.

Successfully estimated effect of uncertainties, prior information on observable data (measured TLD dose) of organ doses from CT exposure.

Provides the platform for future research directions focusing on further refining the Bayesian hierarchical model and exploring its applicability in clinical practice.

Enables specific organs dose estimation of patients to ensure patient protection.