Semantic redundancy-aware implicit neural compression for multidimensional biomedical image data

With the rapid development in advanced imaging techniques, massive image data have been acquired for various biomedical applications, posing significant challenges to their efficient storage, transmission, and sharing. Classical model-or learning-based compression algorithms are optimized for specific dimensional data and neglect the semantic redundancy in multidimensional biomedical data, resulting limited compression performance. Here, we propose a Semantic redundancy based Implicit Neural Compression guided with Saliency map (SINCS) approach which achieves high-performance compression of various types of multi-dimensional biomedical images. Based on the first-proved semantic redundancy of biomedical data in the implicit neural function domain, we accomplished saliency-guided implicit neural compression, thereby notably improving the compression efficiency for large-scale image data in arbitrary dimensions. We have demonstrated that SINCS surpasses the alternative compression approaches in terms of image quality, compression ratio, and structure fidelity. Moreover, with using weight transfer and residual entropy coding strategies, SINCS improves compression speed while maintaining high-quality compression. It yields near-lossless compression with over 2000-fold compression ratio on 2D, 2D-T, 3D, 4D biomedical images of diverse targets ranging from single virus to entire human organs, and ensures reliable downstream tasks, such as object segmentation and quantitative analyses, to be conducted at high efficiency.