AI-ready image compression for whole-slide images using bottleneck tensor compression

Digitized whole-slide images (WSI) are a major biomedical data modality, which is increasing exponentially and enabling artificial intelligence (AI) approaches to biomedical research questions. In this work, we use a state-of-the-art model to compress WSIs as three-dimensional arrays (tensors) that are amenable to storage in Next Generation File Formats. This approach minimizes data storage and allows the reuse of compressed data from arbitrary sub-images for downstream AI models.