基于多层次协同稀疏回归模型侧脑室分割方法

A novel lateral ventricles segmentation method based on multiscale collaborative sparse regression

稀疏学习理论是高光谱解混中的有力工具。由Ioradche等人提出的协同稀疏回归模型[1]利用丰度矩阵的行稀疏特性,对丰度矩阵施加协同稀疏,从而优化解混的结果。本文受到高光谱解混理论的启发,将协同稀疏回归引入到脑室分割中。为了克服传统的协同稀疏回归方法仅考虑噪声误差而忽视稀疏粗差的缺点,本文提出了一种新的基于多层次协同稀疏回归模型侧脑室分割方法,从而进一步提高医学图像的分割精度。该方法将输入的侧脑室形状正则化为形状库中训练形状的稀疏线性组合,并且使用协同稀疏性来描述侧脑室形状库中的行稀疏性。最后采用多层次分割优化策略[2],分割结果将按照由粗到细的方法演化。实验结果表明本文新提出的方法的有效性和实用性。

Sparse Learning Theory is one of the powerful tools for hyperspectral unmixing.The collaborative sparse regression model proposed by Ioradche et al.[1]exploits the row-sparse characteristic of the fractional abundances to impose the collaborative sparsity on the fractional abundances,which impoves the unmixing results.Inspired by Hyperspectral Unmixing Theory,the paper introduces the collaborative sparse regression model into lateral ventricles segmentation.In order to overcome the shortcomings of traditional collaborative sparse regression methods which only pay attention to the reconstruction noise error and neglect the sparse error,the paper proposes a novel lateral ventricles segmentation method based on multiscale collaborative sparse regression to further improve the accuracy of lateral ventricles segmentation.This method regards the input shape of lateral ventricles as a sparse linear combination of training shapes in a shape repository,and depicts the the row-sparse characteristic of the shape repository of lateral ventricles with the collaborative sparsity.Finally,a multiscale segmentation optimization strategy is developed[2],where the input shape is deformed in a coarse-to-fine manner.The experimental result is provided to illustrate the effectiveness and applicability of the novel method.