Using adaptive neuro-fuzzy inference system technique for crosstalk correction in simultaneous ~(99m)Tc/~(201)Tl SPECT imaging: A Monte Carlo simulation study

摘要

This work presents a simulation based study by Monte Carlo which uses two adaptive neuro-fuzzy inference systems (ANFIS) for cross talk compensation of simultaneous ~(99m)Tc/~(201)Tl dual-radioisotope SPECT imaging. We have compared two neuro-fuzzy systems based on fuzzy c-means (FCM) and subtractive (SUB) clustering. Our approach incorporates eight energy-windows image acquisition from 28 keV to 156 keV and two main photo peaks of ~(201)Tl (77 ± 10% keV) and ~(99m)Tc (140 + 10% keV). The Geant4 application in emission tomography (GATE) is used as a Monte Carlo simulator for three cylindrical and a NURBS Based Cardiac Torso (NCAT) phantom study. Three separate acquisitions including two single-isotopes and one dual isotope were performed in this study. Cross talk and scatter corrected projections are reconstructed by an iterative ordered subsets expectation maximization (OSEM) algorithm which models the non-uniform attenuation in the projection/back-projection. ANFIS-FCM/SUB structures are tuned to create three to sixteen fuzzy rules for modeling the photon cross-talk of the two radioisotopes. Applying seven to nine fuzzy rules leads to a total improvement of the contrast and the bias comparatively. It is found that there is an out performance for the ANFIS-FCM due to its acceleration and accurate results.