LAUSR

Novel breast dosimetry applications require the knowledge of incident X-ray spectra, in mammography energy range [1] . Furthermore, the knowledge of the spectra transmitted through breast tissue is important in the phantom development context. A vast number of MC codes are used as models to generate diagnostic X-ray spectra; however, comparisons of those to experimentally measured X-ray spectra (validations) must be done to assure modelled spectra are representative of real X-ray beams. For these comparisons to be reliable, accurate uncertainty estimation of the counts in the spectra channels is required. This work’s objective was to accurately assess the uncertainties associated to the counts of incident and transmitted mammography MC modelled spectra. In this work, the PENELOPE MC code was used to simulate a CdTe spectrometer’s detection of mammography X-ray spectra, and the attenuation of these spectra after transmission through commercial breast phantoms. Four anode-filter combinations, three commercial breast phantoms and several attenuation thicknesses were used. Altogether, 108 spectra were simulated. The measured spectra emitted by each X-ray generators were used as the MC simulations input spectra, after escape and efficiency distortion effects were corrected by the stripping method. The uncertainty of the modelled spectra’s channels was estimated through MC uncertainty propagation method. For each condition, 25 simulations were performed, varying the input parameters (input spectra and phantom composition) according to their probability density functions. The simulated spectra validation was done through the assessment of the mean weighted squared residuals (MWSR). Except for model’s inaccuracies, good estimation of the uncertainties yields a MWSR equal to 1, which is this quantity’s expected value. The mean of the 108 calculated MWSRs yielded 3.76(03) when only Poisson’s uncertainty was used. Therefore, we conclude Poisson’s uncertainty underestimates the uncertainty of the spectra channels’ counts. The estimation we performed demonstrated better accuracy, it yielded a mean MWSR of 1.17(01).