Scatter Estimation by Convolution Filtering: Application to Digital Subtraction Angiography.
This dissertation describes investigation of the effects of iodine beam hardening and scatter on the response linearity of a digital fluoroscopic x-ray imaging system to iodine concentration. A DSA (digital subtraction angiography) system was investigated. An upward shift in the average x-ray beam energy --beam hardening--caused by differential x-ray absorption by the iodine contrast was investigated both in the absence and presence of scatter. Scatter includes x-ray scatter within the patient and light and electron scatter within the image detector. In the scatter-free case, a computer model and experimental measurements showed good agreement. If C, the DSA image contrast--the image signal--was fit to a power law of the form C = A(pt)('Y), where pt is the iodine areal density (product of iodine density and thickness in mg/cm('2)), a value of Y = 0.93 gave the best fit for the scatter free case. Results from similar experiments performed in the presence of scatter gave a value of Y = 0.72 as the best fit. Thus iodine beam hardening effects were found to be secondary to the effects of scatter in producing a nonlinear response of the DSA imaging system to varying iodine concentration. Two techniques to estimate the regional scatter distribution in DSA images were investigated. Both utilized local scatter estimates obtained from small lead beam-stops placed in the x-ray beam. A least-squares fit to the beam -stop values yielded approximately a 15% rms error with respect to the beam-stop values over the entire image. A convolution-filtering approach proved most successful in estimating the scatter. Investigation of more than 175 convolution kernels, applied to ten imaging situations, using three imaging fields-of-view showed that a kernel of the form exp(-ax) in each image dimension, with a FWHM (full width at half maximum) of 50 to 150 pixels best reproduced the scatter disributions within these images, with an rms percentage error from 4 to 8%. Such a kernel with a FWHM of 75 pixels in each dimension applied to all the images resulted in an average rms percentage error of 6.6%. Estimating the scatter in an image containing iodine using the image beam-stop data from an iodine-free image resulted in less than 1% increase in the error, averaged over all image studies.
- Pub Date:
- Physics: Radiation