Improved image derived input functions using Iterative deconvolution and HYPR denoising

Objectives Ten subjects, each undergoing three [¹⁵O]H₂O PET scans (2 baseline: test and retest, 1 after hypercapnia) were included in this study.

Methods The following steps were used to generate each IDIF: (a) to accurately draw the region of interest (ROI), a gradient image was generated from an early frame of the dynamic scan highlighting the vessel walls; (b) a ROI (lumen of carotid artery) was drawn manually on this gradient image; (c) to reduce partial volume effects, the Van Cittert iterative deconvolution method (IDM) with 8 iterations and FWHM of 6.75 mm was used; (d) to reduce noise, HYPR with a moving composite (3 parts: Influx, efflux and equilibrium phases of the blood input) was used. Parametric CBF images were generated using the original dynamic [¹⁵O]H₂O images in combination with (1) measured on-line arterial input functions and (2) IDIFS derived from both original images and those processed as indicated above. Regional parametric test, retest and hypercapnia CBF values were used to test accuracy and precision when using image derived input functions.

Results Average absolute test-retest variabilities of 7±5 (measured arterial input function), 20±16 (original IDIF) and 13±11% (IDM and HYPR processed IDIF) were obtained for whole brain grey matter CBF. Biases of 82±30, 20±14% in CBF were found using original and improved IDIs, respectively, as compared with CBF values obtained with measured arterial input functions.

Conclusions IDIFs obtained using iterative deconvolution in combination with HYPR denoising resulted in CBF repeatability and accuracy than were substantially better than those for original IDIFs. Nevertheless, CBF values were, on average, still 20% higher than those obtained using measured arterial input functions. The reduced precision and accuracy are possibly caused by higher noise levels in the IDIFs at later time points (>300 s p.i.). Further studies will be performed using shorter scan durations (<300 s) which might improve repeatability and accuracy of CBF based on IDIFs. References: 1. Teo BK, Seo Y, Bacharach SL, Carrasquillo JA, Libutti SK, Shukla H, et al. Partial-volume correction in PET: validation of an iterative postreconstruction method with phantom and patient data. J Nucl Med 2007 May;48(5):802-810. 2. Lucy L. An iterative technique for the rectification of observed distributions. 79 ed. Astron. J.: 1974, pp. 745-765. 3. Christian BT, Vandehey NT, Floberg JM, Mistretta CA. Dynamic PET denoising with HYPR processing. J Nucl Med. 2010;51(7):1147-54.