Brain Tissue Microstructure Parameters Estimation Using Iterative Soft Thresholding Network

Neurite orientation dispersion and density imaging(NODDI)is a new diffusion magnetic resonance microstructure imaging technique that can detect the complexity of the neurite microstructure to reflect the morphological information of nerve fibers.However,the traditional NODDI fitting algorithm has a high computational complexity and noise sensitivity,which limits wide clinical application.At present,deep learning has been successfully applied in the field of medical imaging.Studies have shown the method of microstructure parameter estimation based on deep learning can improve calculation accuracy and speed of fitting,which is of great significance to clinical medical diagnosis.Aiming at the shortcomings of traditional NODDI fitting method such as slow calculation speed and susceptibility to noise interference,this thesis proposes a deep learning-based microstructure parameter estimation method,ISTA-Net,taking into account a large amount of high-quality diffuse magnetic resonance data for neural networks training,this article generates two types of simulation data sets: based on the NODDI-Watson model single-cell simulation and whole-brain simulation data sets.The simulation process introduces the effects of noise and motion artifacts,which is convenient to provide a reference standard to objectively measure differences from many aspects.The performance of the fitting algorithm.The method in this thesis,ISTA-Net,sparsely encodes the diffusion signal in the angular and spatial domains,and then de-folds the soft threshold iterative algorithm to build a network to solve the convex optimization problem of l1 norm,and then calculates the mapping of sparse representation to microstructure indicators.The combination of traditional threshold iterative algorithm and deep learning makes the network framework interpretable and reduces the computational complexity.It can still achieve accurate parameter estimation under the under-sampling diffusion gradient.Finally,this thesis verifies the performance of the proposed algorithm from multiple aspects on the simulation data set and the public data set.The image quality is measured by indicators such as mean square error,structural similarity,and peak signal-to-noise ratio.Experimental results show that the algorithm can still achieve accurate estimation of brain tissue microstructure under the influence of reduced diffusion gradients or noise.The ISTA-Net proposed in this paper has certain advantages over other algorithms in terms of computational speed and accuracy,which is helpful Because NODDI technology will be widely used in clinical and scientific research in the future.