TRANSFER-FUNCTION ANALYSIS OF UFCT MYOCARDIAL TIME-DENSITY CURVES BY TIME-VARYING RECURSIVE LEAST-SQUARES ANALYSIS

Techniques which assume linear, time-invariant systems have been used to characterize indicator dilution pairs. As a basis for fully describ ing the relation between left ventricular (LV) and myocardial (MYC) ti me-density curves, produced by an intravenous constrast medium as meas ured by ultrafast CT, the assumption of time invariance was tested usi ng recursive least squares regression and CUSUM, a test for time varia bility of regression parameters. Using data from anesthetized dogs wit h concomitant microsphere information, constant and time-varying regre ssion models, MYC(t) = b(t)LV(t-1), were generated from time-density c urves of flows from two groups: Group 1 (MBF < 2 ml/min/gm, n=11) and Group 2 (MBF > 2 ml/min/gm, n=10). The time-varying regression models had reduced root mean square error: 0.6 +/- 1.1 and 0.5 +/- 0.8 versus 7.3 +/- 3.5 and 4.1 +/- 1.6 for Groups 1 and 2, respectively. Signifi cant time variability p < 0.05) by CUSUM was found in 9/11 Group 1 mod els and 7/10 Group 2 models. Myocardial blood volume was estimated as the average value of b(t) over the rising portion of the LV curve. Myo cardial blood flow was then calculated as myocardial blood volume divi ded by coronary transit time, determined from gamma variate fits of th e LV and scaled, shifted LV curve, with excellent results over a wide range of flows (r = 0.93, y = 0.92x + 0.28, range of 0.4 to 6.7 ml/min /gm). These results show that measurements of increased myocardial blo od flow are possible with an intravenous contrast media, and that move ment of contrast medium from intravascular space to extravascular spac e occurs during the course of the contrast medium's first pass.