The effects of noise in cardiac diffusion tensor imaging and the benefits of averaging complex data

There is growing interest in cardiac diffusion tensor imaging (cDTI), but, unlike other diffusion MRI applications, there has been little investigation of the effects of noise on the parameters typically derived. One method of mitigating noise floor effects when there are multiple image averages, as in cDTI, is to average the complex rather than the magnitude data, but the phase contains contributions from bulk motion, which must be removed first. The effects of noise on the mean diffusivity (MD), fractional anisotropy (FA), helical angle (HA) and absolute secondary eigenvector angle (E2A) were simulated with various diffusion weightings (b values). The effect of averaging complex versus magnitude images was investigated. In vivo cDTI was performed in 10 healthy subjects with b = 500, 1000, 1500 and 2000 s/mm(2). A technique for removing the motion-induced component of the image phase present in vivo was implemented by subtracting a low-resolution copy of the phase from the original images before averaging the complex images. MD, FA, E2A and the transmural gradient in HA were compared for un-averaged, magnitude- and complex-averaged reconstructions. Simulations demonstrated an over-estimation of FA and MD at low b values and an under-estimation at high b values. The transition is relatively signal-to-noise ratio (SNR) independent and occurs at a higher b value for FA (b = 1000-1250 s/mm(2)) than MD (b ≈ 250 s/mm(2)). E2A is under-estimated at low and high b values with a transition at b ≈ 1000 s/mm(2), whereas the bias in HA is comparatively small. The under-estimation of FA and MD at high b values is caused by noise floor effects, which can be mitigated by averaging the complex data. Understanding the parameters of interest and the effects of noise informs the selection of the optimal b values. When complex data are available, they should be used to maximise the benefit from the acquisition of multiple averages. The combination of complex data is also a valuable step towards segmented acquisitions.