Chemometrics Group , Universitat de Barcelona , Diagonal, 645 , 08028 Barcelona , Spain.
IDAEA-CSIC , 08034 Barcelona , Spain.
Anal Chem. 2018 Jun 5;90(11):6757-6765. doi: 10.1021/acs.analchem.8b00630. Epub 2018 May 7.
Data fusion of different imaging techniques allows a comprehensive description of chemical and biological systems. Yet, joining images acquired with different spectroscopic platforms is complex because of the different sample orientation and image spatial resolution. Whereas matching sample orientation is often solved by performing suitable affine transformations of rotation, translation, and scaling among images, the main difficulty in image fusion is preserving the spatial detail of the highest spatial resolution image during multitechnique image analysis. In this work, a special variant of the unmixing algorithm Multivariate Curve Resolution Alternating Least Squares (MCR-ALS) for incomplete multisets is proposed to provide a solution for this kind of problem. This algorithm allows analyzing simultaneously images collected with different spectroscopic platforms without losing spatial resolution and ensuring spatial coherence among the images treated. The incomplete multiset structure concatenates images of the two platforms at the lowest spatial resolution with the image acquired with the highest spatial resolution. As a result, the constituents of the sample analyzed are defined by a single set of distribution maps, common to all platforms used and with the highest spatial resolution, and their related extended spectral signatures, covering the signals provided by each of the fused techniques. We demonstrate the potential of the new variant of MCR-ALS for multitechnique analysis on three case studies: (i) a model example of MIR and Raman images of pharmaceutical mixture, (ii) FT-IR and Raman images of palatine tonsil tissue, and (iii) mass spectrometry and Raman images of bean tissue.
不同成像技术的数据融合可实现对化学和生物系统的全面描述。然而，由于不同的光谱平台获取的图像具有不同的样本方向和空间分辨率，因此将这些图像进行融合是非常复杂的。虽然通过对图像进行适当的旋转、平移和缩放的仿射变换，可以解决样本方向的匹配问题，但在多技术图像分析中，图像融合的主要难点在于如何在保留最高空间分辨率图像的空间细节的同时，将多个技术的图像进行融合。在这项工作中，提出了一种特殊的不完全多集解混算法——多变量曲线分辨交替最小二乘法（MCR-ALS），以解决此类问题。该算法允许在不损失空间分辨率的情况下，同时分析来自不同光谱平台的图像，并确保处理后的图像具有空间一致性。这种不完全多集结构将两个平台的低空间分辨率图像与最高空间分辨率的图像进行拼接。结果，样品的组成成分由一组分布图谱定义，这些图谱是所有使用平台的公共图谱，且具有最高的空间分辨率，同时还包含每个融合技术的扩展光谱特征。我们通过三个案例研究来证明这种 MCR-ALS 新变体在多技术分析中的潜力：（i） 药物混合物的 MIR 和拉曼图像模型示例，（ii） 腭扁桃体组织的 FT-IR 和拉曼图像，以及（iii） 豆组织的质谱和拉曼图像。
