An iterative algorithm for computing gradient directions for white matter fascicles detection in brain MRI

This paper proposes a new iterative algorithm for computing gradient directions (GD) to reconstruct the brain's white matter fascicles. In particular, the proposed algorithm extensively overcomes the limitations of existing approaches like Uniform Gradient Directions and Adaptive Gradient Directions (AGD) for this task. The proposed algorithm uses the AGD approach to have a coarse estimation of the fibers in the initial step, and then a refinement is done using an iterative strategy. We begin with GD distributed uniformly inside a grid of bigger size and larger spacing between the points. Both (grid size and spacing between the points) reduce iteratively. The proposed algorithm has higher chance of capturing the fibers' actual position within the grid at each iteration. Hence, the solution tends to the actual position of fiber in each iteration, leading to a better estimation of fiber orientations. Multiple artificial simulations and real dataset experiments on the human brain and optic chiasm of a rat's brain are performed. The excellent performance of the proposed algorithm at different noises ensures stability and robustness. Hence, after processing the MRI data, the proposed algorithm can accurately reflect the ground truth of white matter fascicles connections in reconstructed images. The proposed algorithm helps resolve the structural complexities of the brain caused due to presence of crossing fascicles to a great extent.