Detecting latent spatial patterns in mass spectrometry brain imaging data via Bayesian mixtures

Mass spectrometry methods can record biomolecule abundance for a broad set of molecular masses given a sample of a specific biological tissue. In particular, the MALDI-MSI technique produces imaging data where, for each pixel, a mass spectrum is recorded. There is the urge to rely on suited statistical methods to model these data, fully addressing their morphological characteristics. Here, we investigate the use of Bayesian mixture models to segment these real biomedical images. We aim to detect groups of pixels that present similar patterns to extract interesting insights, such as anomalies that one cannot capture from the original pictures. This task is particularly challenging given the high dimensionality of the data and the spatial correlation among pixels. To account for the spatial nature of the dataset, we rely on Hidden Markov Random Fields