NIR-VIS multispectral fused imaging with extended depth-of-field for high-throughput virtual staining

Accurate histological analysis plays an irreplaceable role in disease diagnosis. Although virtual staining has alleviated some of the costly and tedious steps involved in histochemical staining processes, most methods employ low-throughput inputs, potentially limiting accurate reconstruction of tissue architecture. In this paper, a high-throughput virtual staining framework is proposed to achieve robust characterization of tissue structural features with extended depth-of-field by integrating information across visible (VIS) and near-infrared (NIR) ranges from 10-µm-thick tissue sections. A multispectral microscopic imaging system is constructed to acquire multispectral information. HSSNet is established to efficiently extract high-dimensional spectral features, achieving accurate mapping from plain multispectral transmittances to stained RGB images. Experimental results on liver cancer across multiple differentiation grades demonstrate that our method maintains high color fidelity and reliably reconstructs the differentiation-dependent tissue structural features, providing a promising paradigm for extending virtual staining into histopathological assessment of heterogeneous tumors.