Asymmetric Fiber Orientation Distribution Estimation and Response Function Calibration with an Unsupervised Deconvolution Network

Motivation: Fiber orientation distribution function (fODF) and response function (RF) influence each other from diffusion MRI. To capture more detailed fiber patterns, estimating asymmetric fODFs (a-fODFs) and accurate RF helps further exploration on white matter microstructure. Goal(s): Utilize the spatial information and RF calibration to estimate a-FODs and RF synchronously. Approach: We construct an unsupervised spherical deconvolution network to estimate the a-fODFs, and apply it within a recursive framework to calibrate both RF and a-fODFs simultaneously. Results: The fiber connectivity from a-fODFs shows stronger correlations with the labeled neuron connectivity from neuronal tracing data, and exhibits great robustness on test-retest data. Impact: Our method conducts the unsupervised estimation of a-fODFs and RF without the histological data. Compared with symmetric fODFs, the a-fODFs capture complex fiber patterns, providing a broader perspective for downstream tractography and white matter microstructure analysis.