김재민 (서울대학교)

이재규 (서울대학교)

김재민 (서울대학교)

이재규 (서울대학교)

Mass spectrometry imaging (MSI) is a powerful technique for biomedical analysis, providing spatially resolved distributions of metabolites, lipids, and proteins, without the need for staining or labeling. However, a major limitation of MSI is its inherently low spatial resolution, which hinders the precise visualization of molecular compositions. While many attempts have been made using deep learning-based Image enhancement techniques, their effectiveness is limited by the difficulty of acquiring high-resolution MSI images. To overcome this challenge, we propose a Self-Similarity-based Super-Resolution MSI (SSS-MSI) method. This approach leverages the inherent self-similarity within individual MSI images to enhance the spatial resolution of MSI images. Unlike conventional deep learning methods, SSS-MSI achieves high-fidelity super-resolution with only 30 MSI images, significantly reducing data dependency. This method requires a minimal number of datasets for fine-tuning on unseen tissues to achieve the great generalization. Additionally, SSS-MSI converges faster during iterative trainings compared to existing methods, while maintaining superior perceptual and structural fidelity. By enhancing the spatial resolution of MSI images to the single-cell level, SSS-MSI facilitates uncovering spatial cellular heterogeneity, which is critical for understanding tissue organization and disease progression, thereby opening new avenues for improved MSI-based biomedical applications.