Two-Tensor Residual Bootstrapping on Classified Tensor Morphologies

Nagulan Ratnarajah¹, Andy Simmons Simmons², Ali Hojjatoleslami¹

A fast and clinically feasible two-tensor residual-bootstrapping algorithm is applied to classified tensor-morphologies for estimating uncertainty in fibre-orientation and probabilistic-tractography. The classification of tensor-morphologies allows information about the true tensor morphology to be considered when selecting the most appropriate bootstrap procedure to use. We use a constrained two-tensor model for the planar voxels, so MRI acquisition times can be greatly reduced when compared to other multi-tensor approaches which will enable widespread clinical use. Based on our experimental evaluations, unlike previous bootstrap methods with other multi-fibre approaches our algorithm shows improved computational efficiency and accurately reconstructs fibre paths and recovers complex fibre configurations.