逐次近似再構成法における分解能・統計雑音および最適逐次回数の基礎的検討

概要

論文の詳細を見る
We evaluated the optimal number of iterations on maximum likelihood-expectation maximization (MLEM) reconstruction algorithm in SPECT images using four methods: full width at half maximum (FWHM), statistical noise, edge recovery of image (recovery coefficient; RC), and normalized mean squared error (NMSE). We acquired SPECT images using 99m-Tc as a signal source for three phantoms: a line source phantom, uniform cylinder, and hot rod phantom. The line source phantom was used to measure reconstructed image resolution; the uniform cylinder phantom was used to evaluate statistical noise and image recovery; and the hot rod phantom was used to evaluate the NMSE method. We varied acquisition time as follows: 300 sec/step, 30 sec/step, and 5 sec/step, and numbers of iterations of ten times and more. Statistical noise converged early at 50 iterations for acquisition time for 300 sec/step. The RC value neared 1.0 for 300 sec/step, but remained at 0.7 for 30 sec/step and 5 sec/step. The optimal number of iterations was 50 for images obtained with the high frequency ingredient with 30 sec/step and 30 for images obtained with low frequency. The number of iterations should be changed according to acquisition time and the frequency ingredient contained in images.
社団法人日本放射線技術学会の論文
2001-11-20