A New CT Perfusion Analysis Algorithm without Deconvolution: FiTT

Purpose: We propose a new CT perfusion analysis algorithm without deconvolution: Theoretically calculating the time-enhancement curve (TEC) under various perfusion conditions, finding a theoretical TEC that best fits the observed TEC, and using the theoretical TEC’s perfusion parameters as the estimations for the observation point (FiTT). Method: The FiTT analysis procedure was as follows: First, the TEC of the arterial input function (AIF) was fitted to the gamma distribution function. Next, we defined the residual functions (R(t)s) that assume various perfusion states of brain tissue, and theoretical brain TECs were calculated by convolution of the AIF and R(t)s. Finally, we determined a theoretical TEC with the least error from the observed brain tissue TEC, using the theoretical TEC’s perfusion parameters as the estimations for the observation point. The estimation accuracy of FiTT was comparing an academic analyzer (bSVD algorithm) and a commercial analyzer (Bayesian algorithm) using a digital phantom. The verification items were visual evaluation of parametric maps, the relationship between ground truth and estimates, the effect of R(t) shape, the effect of AIF type, and the effect of source image noise. Results: Parametric map findings: For the academic analyzer, Tmax was affected by mean transit time (MTT); for the commercial analyzer, MTT was modulated by delay; for FiTT, CBV was slightly affected by MTT. The global characteristics and effect of the R(t) shape: The academic analyzer underestimated MTT, and tracer delay varied; the commercial analyzer had a positive bias for cerebral blood volume (CBV); and for FiTT, CBV and tracer delay estimated close to ground truth, cerebral blood flow (CBF) and MTT were affected by the R(t) shape. The effect of AIF type: The academic analyzer was independent; for the commercial analyzer, CBF, CBV, and tracer-delay were affected; for FiTT, CBF and CBV were affected. The effect of source image noise on parametric map noise: The academic analyzer was affected slightly; the commercial analyzer was unaffected; for FiTT, CBF map was affected. The effect of source image noise on the estimation bias: All analyzers were independent. The effect of the source image noise on the correlation coefficients (ground truth and estimates): None of the analyzers showed obvious proportional relationships. Discussion and conclusions: The three analyzers were affected by one or more of noise, AIF type, and R(t) shape. Therefore, the estimates were overestimated, underestimated, biased, varied, and interacted. FiTT depended on R(t) shape and parametric map noise was affected by source image noise; however, estimates were close to ground truth and showed good linearity in many cases. Because clinical cases were not evaluated, the clinical behavior of FiTT is unknown. Within the scope of this study, we conclude that the estimation accuracy of FiTT is comparable to that of the academic and the commercial analyzer.