Abstract
160
Objectives The objective of this study was to acquire time-activity-curves (TACs) by wavelet packets based sub-band decomposition independent component analysis (WP-based SBICA) and automatic selection of the whole blood time-activity-curve (wTAC) using an artificial neural network (ANN).
Methods There were two types of materials in the experiments: (1) 100 sets of simulated dynamic rat images (2) dynamic images of six real Sprague-Dawley rats. First, each TAC was estimated by WP-based SBICA. Second, a wTAC was selected using an ANN. We also used FastICA estimated wTAC to compare with the results of our method using the normalized root mean square error (NRMSE) and error of area under curve (EAUC).
Results In the simulated study the averaged NRMSEs for the two methods were 0.15±0.02 and 0.24±0.02 for WP-based SBICA and FastICA, respectively. The averaged EAUCs were 0.03±0.02 and 0.07±0.02 for WP-based SBICA and FastICA, respectively. In the realistic rat study, the averaged NRMSEs for the two methods were 0.25±0.05 and 0.34±0.10 for WPICA and FastICA, respectively. The averaged EAUCs were 0.11±0.01 and 0.23±0.19 for WP-based SBICA and FastICA, respectively.
Conclusions Our results show that the accuracy of estimation of the wTAC by using WP-based SBICA and an ANN outperforms that obtained by FastICA.
Research Support This work was funded by NSC98-3113-B-010-012 through the support of MAGIC core