Interpretable dimensionality reduction and classification of mass spectrometry imaging data in a visceral pain model via non-negative matrix factorization

Abstract

Mass spectrometry imaging (MSI) is a powerful scientific tool for understanding the spatial distribution of biochemical compounds in tissue structures. MSI data analysis presents problems due to the large file sizes and computational resource requirements and also due to the complexity of interpreting the raw spectral data. Dimensionality reduction techniques that address the first issue do not necessarily result in readily interpretable features. In this paper, we present non-negative matrix factorization (NMF) as a dimensionality reduction algorithm that reduces the size of MSI datasets by three orders of magnitude with limited loss of information, yielding spatial and spectral components with meaningful correlation to tissue structure. This analysis is demonstrated on an MSI dataset for an animal model of comorbid visceral pain hypersensitivity (CPH). The significant findings are: 1) High-dimensional MSI data (~ 100,000 ions per pixel) was reduced to 20 spectral NMF components with < 20% loss in reconstruction accuracy. 2) Spatial NMF components are reproducible and correlate well with H&E-stained tissue images. 3) Spatial NMF components may be used to provide images with enhanced specificity for different tissue types. 4) Small patches of NMF data (i.e., 20 spatial NMF components over 15 x 15 pixels) provide an accuracy of ~87% in classifying CPH vs naive control subjects. This paper presents novel methodologies for data augmentation to support classification, ranking of features according to their contribution to classification, and image registration to support tissue-specific imaging.