IDAS Seminars

29.11.2019 Data Science Forum: Bayes linear methodology for analysis of medical X-Ray images

by Benjamin Lopex (Mathematical Sciences (Durham))

218 (OC)




An X-ray image is produced by recording the energy deposited at a detector from photons directed through an entity. The photons can either i) traverse the object unperturbed, ii) be absorbed or iii) scatter. Scatter is considered a form of spatially correlated error which degrades image quality by adding a deformed projection of the entity onto the image. One approach to mathematically correcting scatter is to use a simulator that models the interactions between the X-rays and the entity. However, i) these simulators are expensive to evaluate, ii) scatter depends on the chemical composition of the entity which is often unknown and iii) there will be discrepancy between the simulator and the X-ray system it purports to represent. In this talk, a Bayes Linear methodology addressing these issues will be presented.