A non-negative fast multiplicative algorithm in 3D scatter-compensated SPET reconstruction

Abstract

Single-photon emission tomographic (SPET) reconstruction can be improved, especially for noisy images, by using the iterative expectation-maximization of the maximum-likelihood (EM-ML) algorithm. Its application to clinical routine is, however, hampered by the high number of iterations necessary to achieve acceptable results. Therefore various methods have been developed to accelerate the EM-ML algorithm. In this paper a new accelerated EM-ML-like multiplicative algorithm is proposed for SPET reconstruction. Contrary to some other accelerating methods, it preserves two of the most important properties of the EM-ML, namely pixel positivity inside the patient body and null activity outside. The convergence speed is improved by a factor which can reach 100 in high spatial frequency or low count regions. Good estimates in the low count region are obtained without any smoothing, even at typical routine clinical count rates. The algorithm used in conjunction with the 3D effective one scatter path model provides high-quality SPET images and accurate quantitation.

Inadequate detection of accessory spleens and splenosis with laparoscopic splenectomy A shortcoming of the laparoscopic approach in hematologic diseases

“…Because of the proximity of high liver activity, we found that iterative reconstruction [37] provided better results than standard filtered back projection. Underdetection of occult AS during laparoscopic splenectomy was the cause of these residual spots in at least one of our patients.…”

“…Reconstruction was performed by means of an iterative Љfalse likelihood algorithmЉ [37] with scatter correction [36]. This algorithm is an accelerated version of the Љexpectation of maximum likelihoodЉ (EML) method, which preserves the pixel positivity inside the body and null activity outside.…”

Diagnostic and Therapeutic Potential of Amyloid-Reactive IgG Antibodies Contained in Human Sera

“…Imaging data were recorded using a micro single photon emission computed tomography (SPECT) instrument capable of high-level spatial resolution to ϳ1.7 mm (18). For each study, 60 projections were acquired at 6°intervals, and the images were reconstructed using an implementation of the EM-ML algorithm (19). After completion of the SPECT study, animals were placed in the high-resolution (to 50 m) micro-computed tomography (CT) apparatus, and 180 projections were collected.…”

Median root prior and ordered subsets in Bayesian image reconstruction of single-photon emission tomography

“…MRP generates quantitatively stable solutions with respect to the number of iterations [1]. This is a very interesting property since it is well known that the increase in the iteration number tends to generate less quantitatively accurate solutions in the MLEM process [1][2][3]. Among the major advantage of MRP, two are worth mentionning.…”

“…For application in clinical routine, the use of an accelerating procedure is highly desirable. This is for example particularly true in low count density regions where a large number of iterations is required [3]. An accelerating procedure that would limit the noise generated by the numerous iterations would therefore be very appreciable.…”