Efficient Compression of Mass Spectrometry Images via Contrastive Learning-Based Encoding

In this study, we introduce a novel encoding algorithm utilizing contrastive learning to address the substantial data size challenges inherent in mass spectrometry imaging. Our algorithm compresses MSI data into fixed-length vectors, significantly reducing storage requirements while maintaining crucial diagnostic information. Through rigorous testing on data sets, including mouse bladder cross sections and biopsies from patients with Barrett's esophagus, we demonstrate that our method not only reduces the data size but also preserves the essential features for accurate analysis. Segmentation tasks performed on both raw and encoded images using traditional <i>k</i>-means and our proposed iterative <i>k</i>-means algorithm show that the encoded images achieve the same or even higher accuracy than the segmentation on raw images. Finally, reducing the size of images makes it possible to perform t-SNE, a technique intended for frequent use in the field to gain a deeper understanding of measured tissues. However, its application has so far been limited by computational capabilities. The algorithm's code, written in Python, is available on our GitHub page https://github.com/kskrajny/MSI-Segmentation.