Hi all, for those that are interested,
I found the problem to be related to the output of 5ttgen freesurfer. As per this discussion 5ttgen fsl vs. 5ttgen freesurfer and this paper https://www.researchgate.net/publication/342800028_Hybrid_Surface-Volume_Segmentation_for_improved_Anatomically-Constrained_Tractography (courtesy of Robert Smith and Claude Bajada) 5ttgen hsvs provided a much better result.
It was not without it’s problems. I encountered errors in ~10% of my subjects identical to those experienced here: 5TT hsvs fails in some but not all subjects - #5 by NicDC . The solution at the end of the chain proved very helpful.
Many thanks