Often used for the imaging of compounds with short transverse relaxation times (<1 ms), 3D center-out sequences performing broadband excitation on a ramped-up strong gradient provide very short acquisition delays and high bandwidth [1] (Fig. 1). Because of finite transmit-receive deadtimes ($$$\Delta$$$), data located around the k-space center is missed and needs to be recovered. To do so, different approaches, often referred to as “zero echo time” techniques, like algebraic ZTE [2], WASPI [3,4] or PETRA [5] are used today. Typically, deadtimes of 5-60 us are realized to target tissues with T2s of 150-500 us with overall promising results. However, an important but still open question is how the relation of the deadtime and the T2s involved affect image quality. Therefore, we investigate the performance of state-of-the-art zero-TE techniques depending on the deadtime duration. We focus on the case where components are present with ultra-short-T2s of tens of us coming from hardware parts or solid tissues, showing that critical situations can occur for similar values of deadtime and T2.

Sequences: The three techniques use the same basic pulse sequence (Fig. 1) and differ only in the way they recover the data lost during the deadtime (Fig 2). In ZTE, this issue is addressed by radial oversampling and algebraic reconstruction [2] up to gap sizes of 3-4 Nyquist dwells (limited by ill-conditioning) [6], optionally adding k=0 [7]. In PETRA, the lacking centre is measured with single-point imaging (SPI). In WASPI, a second ZTE acquisition is performed at lower gradient strength with optional T2 correction [3] or linear merging in an overlap region [4]. Note that strictly speaking, any method relying on the direct measurement of k0 will have TE=$$$\Delta$$$. Thus only ZTE and WASPI (without k0) are “true” zero-echo-time sequences.

After recovering the missing data, 3D images were reconstructed iteratively with the conjugate gradient method [8] by solving:

$$(E^{H}WDE)v=E^{H}WDm$$

with E=encoding matrix, v=voxel intensities, m=measured data, D=density compensation matrix, W=weighting matrix used for merging in WASPI.

Experiments: First, a stack of rubber erasers (T2* ~=324 us) was imaged with a hard polymer-based surface coil (T2* ~=10 us). Parameters: bandwidth=500 kHz, isotropic resolution=3.75 mm, max. gradient=24.5 mT/m, minimum $$$\Delta$$$=6 us.

Second, a proton-free PTFE-based surface coil was used to image a piece of bovine tibia. Long-lived signals were removed by releasing the bone marrow, leaving mainly bone matrix (T2* ~=10 us) and bound water (T2* ~=160 us). Parameters: bandwidth=250 kHz, isotropic resolution=1.6 mm, max. gradient=29.4 mT/m, minimum $$$\Delta$$$=6 us.

The effect of deadtime on image quality was then investigated by removing data up to 40 us.

1D PSF simulations were performed to investigate the sequences’ performance for ultra-short-T2 signal (Fig. 3). Pure ZTE (without k0) is the only method able to reconstruct the expected ideal Lorentzian lineshape. All other methods diverge from it when T2 approaches $$$\Delta$$$ . PETRA and ZTE with k0 exhibit a narrower center lobe and two side lobes of low amplitude. In WASPI, the main lobe amplitude is lowered while side lobes of increased intensity appear, with variations depending on overlap and use of k0.

Fig 4: The performance observed for the ultra-short-T2 signal of the coil confirms the findings of the PSF simulations. Pure ZTE works well at $$$\Delta$$$=6 us while at $$$\Delta$$$=8 us aliasing of out-of-band signal in the FOV center can be removed by adding k0. PETRA exhibits only minor ringing artifacts at smaller $$$\Delta$$$ while coil signal is suppressed at $$$\Delta$$$=40 us. In WASPI, strong ringing artifacts are observed, at $$$\Delta$$$=40 us even from the longer T2 signal of rubber.

Fig 5: Apart from blurring due to apodization, the ultra-short-T2 components of the bone (mainly collagen) are well reconstructed by ZTE and PETRA at $$$\Delta$$$=12 us. In WASPI, this signal is lost associated with data overlap. At $$$\Delta$$$=40 us, image intensity becomes dominated by bound water which is depicted well only with PETRA.

Compounds with short transverse relaxation times (hundreds of us) can be imaged with high fidelity with all investigated “zero-TE” techniques. However, for deadtimes comparable to the T2s involved, careful consideration is required. ZTE provides maximum short-T2 sensitivity with almost ideal PSF at moderate deadtimes, but may suffer from artifacts due to out-of-band signal at larger deadtimes [6]. In this case, PETRA offers favorable performance, and also enables selecting T2 sensitivity by variation of the deadtime duration. In contrast, WASPI can lead to oscillating PSFs and thus image artifacts, whereupon T2 correction [3] is limited to comparable T2 values.