Structure-aware enhancement of imaging mass spectrometry data for semantic segmentation

Mass Spectrometry Imaging data contains structural information, where similar mass spectra represent the same object. However, due to data contaminations during the measurement, the structural information in the image is in-apparent. We develop a new approach to enhance these structures and then semantically segmenting the given Mass Spectrometry Imaging data by following the enhanced structures. After the pipelined steps of image enhancement, raw segmentation and semantic clustering, meaningful color-coded image segmentation is produced, which greatly captures the main structure of the image and also suppress pixel-wise variations introduced during the measurement. Comparisons show the effectiveness of our pipeline. A biological application based on our enhancement and segmentation shows that our method can be used to identify regions of tissue sections.