本文通过对磁共振信号特征横向驰豫时间T2的分析，提取了标志人类认知功能及情绪发展的生物机制——髓鞘含量的相关信息。本研究基于Bruker4.7T动物磁共振扫描仪，采用Carr-Purcell-Meiboom-Gill （CPMG）射频序列，将Non-Negative Least Squares/Regularized Non-Negative Least Squares (NNLS/rNNLS)模型及双指数模型应用于常用动物模型Rat模型及新型动物模型Marmoset模型，得到了两种动物模型的髓鞘含量分布。除此之外，本研究还通过设置一组模拟数据，加入五种不同等级的噪声，观察rNNLS模型中的正则项在不同信噪比下对于髓鞘含量分布的影响。结果发现，和预期一样，实验所得髓鞘含量分布与动物模型的白质分布呈高度相关。然而，两种动物模型的髓鞘含量分布都不同程度的受到某种伪影的影响。通过进一步分析发现，该伪影是由所采集的横向驰豫衰减曲线导致，在样本中含有相同物质的不同位置取点，两处的回波的差值是不一致的。此现象说明，该扫描仪的CPMG射频序列的各个回波可能受到受激发回波及所选切片外部区域的信号干扰。对于rNNLS模型中的正则项，本研究设置了两个不同的范围，[1 ,1.01 ]和[1.02 ,1.025 ]（文献中常用的范围）。模拟数据的分析结果表明，在模型中加入正则项会使Myelin Water Fraction（MWF,髓鞘含量的一个指标）及Geometric T2（gmT2，刻画T2分布的特征）的值被低估，并且当模型中正则项的权重增大时，这种低估倾向愈发明显。并且随着信噪比降低，MWF及gmT2的准确度也降低，两个指标的分布也愈接近均匀分布。综上所述，NNLS/rNNLS模型适用于从磁共振信号中提取髓鞘含量的相关信息，然而应尽量减小正则项在模型中的权重。

This study investigates the calculation of the amount of myelin which is a biological mechanism indicating human cognitive and emotional development from one of the characteristics of Magnetic Resonance (MR) signal. This study bases on 4.7T Bruker scanner and employed Carr-Purcell-Meiboom-Gill (CPMG) sequence. Non-Negative Least Squares/Regularized Non-Negative Least Squares (NNLS/rNNLS) model and two exponentials model were applied to two different animal models: Rat and Marmoset. Myelin maps were calculated for each animal model. As hypothesized, myelin maps were highly correlated with white matter distributions from animal models. However, both myelin maps were contaminated by unknown noise. After further investigation, we found that the noise was resulted from CPMG sequence. The echoes of the sequence could possibly be influenced by stimulated echoes and signals originating outside the selected slice. Two different ranges were set for the regularizer in rNNLS model, [1 ,1.01 ] and [1.02 ,1.025 ] (widely used in recent papers). Results from simulation data showed that regularizer would underestimate MWF and gmT2. As the weighting of regularizer increases, the underestimation became worse. Along with the decrease of signal to noise ratio, distributions of MWF and gmT2 would approach uniform distribution. In conclusion, NNLS/rNNLS model is suitable for analyzing amount of myelin from MR signal; nevertheless, it is better to reduce the weighting of regularizer in the model.