Abstract ID: 187 Validation of dual energy CT atomic composition extraction using proton Monte Carlo

Proton Monte Carlo (MC) simulations model the physics processes between protons and medium and yield more accurate dose calculation compared to proton stopping power based pencil beam convolution algorithm. MC simulation is also essential for prompt gamma and proton-induced isotope production for in vivo proton range detection. It requires the tissue element composition to be determined from patient CT images. We propose a new hierarchical clustering based approach in extracting the atomic composition in human tissues from Dual Energy CT (DECT) scans and a method for validation using MC simulations. 70 published human tissues samples were used for the theoretical study of this approach. The tissues were first clustered by their composition similarities using hierarchical clustering. For each tissue cluster, three base tissues were randomly selected. Other tissues in this cluster were the weighted sum of these base tissues, where the weights were derived by fitting the attenuation coefficients calculated from DECT X-ray spectrums. The derived tissue atomic compositions and effective atomic number were compared with the ground truth. Animal tissues were scanned using a DECT scanner. The tissues were then irradiated using a PBS proton beam. A double Gaussian proton source has been modeled in GEANT4. Using these images, source model and extraction method, Monte Carlo simulations will be run mimicking the experimental setup. The soft tissues and bones were clustered into 8 subcategories, with cophenetic correlation 0.895 and inconsistency coefficient (IC) > 1.15 ( IC max = 1.1547 ). The mean and standard deviations of the difference in the derived atomic compositions and the ground truth for H, C, N, O, Ca, P were 0.05% (±0.07%), 1.56% (±2.02%), 0.61% (±0.55%), 1.81% (±2.37%), 0.03% (±0.07%), 0.13% (±0.16%), respectively. The Z eff from derived elemental composition was within 0.04 from the true values. We have investigated a new approach to extract the atomic composition from DECT scans. The accuracy of the extraction was demonstrated for published tissue samples. This approach will be validated using the DECT scans of animal tissues and GEANT4 source model, by comparing the Zeff derived from this method against measurements from proton beam irradiations.