IEEE Trans Med Imaging. 2017 Mar;36(3):745-756. doi: 10.1109/TMI.2016.2631001. Epub 2016 Nov 21.
The fMRI signals are usually filtered before processing and analyzing. This process can result in the loss of information carried by the higher frequency in the low frequency fluctuation. ICA and CCA are two classical methods in fMRI. ICA finds the statistically independent components of the observed data, however these components are usually physiologically uninterpretable without auxiliary procedures. CCA decomposes two sets of data into component pairs in some order, however these components may be mixtures of real signals and noise. In order to obtain statistically independent components and avoid the loss of information in the process of filtering, we propose a mixed model based on ICA and CCA, which does not need to filter the data. It is shown by the experiments that the new model has some advantages compared with the classical ICA and CCA. The components obtained by the new model is statistically independent. The useful information included in the low frequency fluctuation can be preserved. Experiments on synthetic data show satisfying results. As an application, this new model is used to design an algorithm to discriminate the major depressions from normal controls, with encouraging experimental results.
功能磁共振成像（fMRI）信号在处理和分析前通常需要进行滤波。这一过程可能导致低频波动中携带的高频信息丢失。独立成分分析（ICA）和典型相关分析（CCA）是 fMRI 中的两种经典方法。ICA 可以找到观察数据的统计独立成分，但这些成分通常需要借助辅助程序才能具有生理学意义。CCA 可以将两组数据分解成某种顺序的成分对，但这些成分可能是真实信号和噪声的混合。为了获得统计独立成分并避免在滤波过程中丢失信息，我们提出了一种基于 ICA 和 CCA 的混合模型，该模型不需要对数据进行滤波。实验表明，与经典的 ICA 和 CCA 相比，新模型具有一些优势。新模型获得的成分在统计上是独立的，可以保留低频波动中包含的有用信息。对合成数据的实验结果令人满意。作为一种应用，该新模型用于设计一种算法，以区分重度抑郁症患者和正常对照组，实验结果令人鼓舞。
