I am assuming you are using MRIcroGL v1.0.20180623 or later. If not, please upgrade.  Hear are some things to consider

• Are you overlaying thresholded statistical maps? If a constant intensity threshold was applied to the whole image you will get much better results by loading the raw unthresholded statistical map and then applying a threshold to the image after it has been warped to the high resolution image. This is because thresholding artificially surrounds the edges of the surviving voxels with darkness, when in fact due to smoothing the neighbors will have similar intensity. For details see the wiki.
• Be aware that SPM normalizes images to have an average sized brain that is roughly aligned to the MNI152 template. In contrast, ANTS and FSL by default warp brains to match the MNI template. Since the MNI152 template is larger than the average brain, brains normalized by SPM are smaller than the MNI152 template. For this reason, I would suggest you use a template generated by SPM - either create one from your population or use the MNI152 template that is shrunken using SPM12's normalization. A sample image of the latter approach is provided on the wiki.
• The Overlays/SmoothImagesWhenLoading option can make images appear a bit more or less jagged. However, this effect is minimized with pre-thresholded statistical maps (see issue 1).
• The Overlays/BackgroundMasksOverlays can hid voxels that are not in the overlay. I would use this with caution, as statistically significant voxels outside the brain are often a sign of poor normalization.