Zhangxuan Hu^{1}, Zhe Zhang^{2}, Yuhui Xiong^{1}, Chun Yuan^{1,3}, and Hua Guo^{1}

Diffusion-weighted imaging can be used to detect orientations of fibers to study human brain connectivity using tractography techniques. Spherical deconvolution based techniques have been widely used for the estimation of fiber orientation distribution (FOD), in which FODs are represented using spherical harmonics coefficients. However, high quality FOD estimation still requires large number of measurements. In this study, a deep neural network based method is proposed to estimate high quality FODs using highly q-space undersampled measurements thus to improve the acquisition efficiency.

**Introduction**

**Methods**

**Results and Discussion**

**Conclusion**

1. Wedeen VJ, Hagmann P, Tseng WY, Reese TG, Weisskoff RM. Mapping complex tissue architecture with diffusion spectrum magnetic resonance imaging. Magnetic resonance in medicine 2005;54(6):1377-1386.

2. Tournier JD, Calamante F, Gadian DG, Connelly A. Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution. NeuroImage 2004;23(3):1176-1185.

3. Tournier JD, Calamante F, Connelly A. Robust determination of the fibre orientation distribution in diffusion MRI: non-negativity constrained super-resolved spherical deconvolution. NeuroImage 2007;35(4):1459-1472.

4. Golkov V, Dosovitskiy A, Sperl JI, Menzel MI, Czisch M, Samann P, Brox T, Cremers D. q-Space Deep Learning: Twelve-Fold Shorter and Model-Free Diffusion MRI Scans. IEEE Trans Med Imaging 2016;35(5):1344-1351.

5. Fan Q, Witzel T, Nummenmaa A, Van Dijk KR, Van Horn JD, Drews MK, Somerville LH, Sheridan MA, Santillana RM, Snyder J, Hedden T, Shaw EE, Hollinshead MO, Renvall V, Zanzonico R, Keil B, Cauley S, Polimeni JR, Tisdall D, Buckner RL, Wedeen VJ, Wald LL, Toga AW, Rosen BR. MGH-USC Human Connectome Project datasets with ultra-high b-value diffusion MRI. NeuroImage 2016;124(Pt B):1108-1114.

Fig.
1 Deep neural network structure, including 9 one-dimensional convolutional layers
(kernel size = 5), 1 flatten layer and 1 dense layer. The inputs of the network
are the normalized signals of 48 measurements. The outputs of the dense layer
are the SH coefficients.

Fig. 2 FODs
calculated using DNN, CSD and the corresponding ground truth in two different
areas.

Fig. 3 MSEs
between the ground truth and the FODs generated by DNN and CSD of four sets of
100 randomly chosen voxels respectively.

Fig. 4
Fiber tracking results of the FODs generated by DNN, CSD and the corresponding ground
truth in two different regions.