Improved visualization of the subthalamic nuclei by reducing susceptibility induced signal losses in T2* weighted multi- gradient-echo images

Introduction T2*-weighted gradient echo (GE) images show a good contrast in iron-rich structures like the subthalamic nuclei (STN) due to microscopic susceptibility induced field gradients with the additional advantage of low specific absorption rate (SAR) exposure [1]. They are therefore useful in Parkinson disease treatment, providing landmarks for exact placement of stimulation electrodes [2, 3]. However, T2*-weighted images are also sensitive to macroscopic field inhomogeneities, resulting in signal losses e.g. in orbitofrontal and temporal brain areas [1, 4, 5] which reduce the anatomical information content. Therefore, on one hand long echo times (TE) are required for achieving a good T2* contrast, on the other hand the extent of signal losses increases with TE. In this work we present an image correction method for multi-echo GE data, consisting of two steps: (1) signal loss correction by evaluation of the phase information and (2) weighted image combination for optimal T2*-contrast and signal-tonoise ratio (SNR) in the deep brain as well as low image degradation in areas affected by macroscopic field inhomogeneities. Materials and Methods Measurements were performed on a 3T whole body MR scanner using a receive-only 8-channel array head coil and the whole body transmit coil. A multi-echo GE sequence (8 echoes, TE = 10/16/22/28/34/40/46/52ms, TR = 900ms, 15 slices, bandwidth = 300Hz/Px, imaging matrix 256*256, in plane resolution 1mm and slice thickness 2mm, total measurement time = 3min 30s) with modulus and phase image output was used. For excitation, an exponentially shaped RF pulse with the time profile A(t) = exp(-4*abs(t)/(P/2)) with the pulse duration P = 2ms and -P/2 <= t <= P/2 was used [6]. Macroscopic gradient maps Gsusc were calculated from the phase difference between the two echoes with the shortest TE and smoothed via convolution with a 5*5 voxel kernel for noise reduction. Processing of magnitude images was performed in 2 steps (Fig. 1): 1.: Images were intensity corrected by pixelwise division by A(Gsusc/Gs*TE), A(t) being the RF-pulse shape and Gs the slice selection gradient [6]. 2.: A combined image was then created as a weighted sum of the intensity corrected images acquired at different TE where the weighting factors were calculated pixelwise for each TE in dependence on Gsusc, in general preferring long TE for achieving good T2* weighting, but short TE in the presence of field inhomogeneities to reduce signal losses. The SNR was measured in STN, white matter (WM) and red nucleus (RN), both in the original, uncorrected images and in the combined images.