Blind separation of cardiac components and extraction of input function from (H2O)-O-15 dynamic myocardial PET using independent component analysis

Citation
Js. Lee et al., Blind separation of cardiac components and extraction of input function from (H2O)-O-15 dynamic myocardial PET using independent component analysis, J NUCL MED, 42(6), 2001, pp. 938-943
Citations number
20
Categorie Soggetti
Radiology ,Nuclear Medicine & Imaging","Medical Research Diagnosis & Treatment
Journal title
JOURNAL OF NUCLEAR MEDICINE
ISSN journal
01615505 → ACNP
Volume
42
Issue
6
Year of publication
2001
Pages
938 - 943
Database
ISI
SICI code
0161-5505(200106)42:6<938:BSOCCA>2.0.ZU;2-L
Abstract
The independent component analysis (ICA) method is suggested to be useful f or separation of the ventricles and the myocardium and for extraction of th e left ventricular input function from the dynamic (H2O)-O-15 myocardial PE T. The ICA-generated input function was validated with the sampling method, and the myocardial blood flow (MBF) calculated with this input function wa s compared with the microsphere results. Methods: We assumed that the eleme ntary activities of the ventricular pools and the myocardium were spatially independent and that the mixture of them composed dynamic PET image frames . The independent components were estimated by recursively minimizing the m utual information (measure of dependence) between the components. The ICA-g enerated input functions were compared with invasively derived arterial blo od samples. Moreover, the regional MBF calculated using the ICA-generated i nput functions and single-compartment model was correlated with the results obtained from the radiolabeled microspheres. Results: The ventricles and t he myocardium were successfully separated in all cases within a short compu tation time (<15 s). The ICA-generated input functions displayed shapes sim ilar to those obtained by arterial sampling except that they had a smoother tail than those obtained by sampling, which meant that ICA removed the sta tistical noise from the time-activity curves. The ICA-generated input funct ion showed a longer time delay of peaks than those obtained by arterial sam pling. MBFs estimated using the ICA-generated input functions ranged from 1 .10 to approximately 2.52 mL/min/g at rest and from 1.69 to approximately 8 .00 mL/min/g after stress and correlated well with those calculated with mi crospheres (y = 0.45 + 0.98x; r = 0.95, P < 0.000). Conclusion: ICA, a rapi d and reliable method for extraction of the pure physiologic components, wa s a valid and useful method for quantification of the regional MBF using (H 2O)-O-15 PET.