Zooming in multi-spectral datacubes using PCA

Imaging mass spectrometry is a technique to determine of which materials a small, physical sample is made. Current feature extraction techniques fail to extract certain small, high resolution characteristics from these multi-spectral datacubes. Causes are a low signal-to-noise ratio, the presence of dominant but uninteresting features, and the huge amount of variables in the dataset. In this paper, we present a zooming technique based on principal component analysis (PCA) to select regions in a datacube for enhanced feature extraction at the highest possible resolution. It enables the selection of spectral and spatial regions at a low resolution and recursively apply PCA to zoom in on interesting, correlated features. This approach is not based on complex and data-specific denoising algorithms. Moreover, it decreases execution time when additional filters have to be applied. The technique utilizes a higher signal-to-noise ratio in the data, without losing the high resolution characteristics. Less interesting and/or dominating features can be excluded in the spectral and spatial dimension. For these reasons, more features can be distinguished and in greater detail. Analysts can zoom into a feature of interest by increasing the resolution.