Green channel-informed collaborative filtering for high-fidelity laser speckle contrast imaging in dual-display laparoscopy

The dual-display laparoscopic imaging system (DLIS), which integrates laser speckle contrast imaging (LSCI) with white-light RGB mode, enhances surgical decision-making during minimally invasive procedures. However, motion artifacts and dynamic perfusion necessitate reduced statistical frames, which increases noise and degrades imaging quality. Although the structure-guided anisotropic approach optimizes local contrast calculation, it fails to suppress global statistical noise. To address this, we propose a green (G) channel-informed collaborative filtering framework, which leverages the hemoglobin-absorption-induced intensity variations in the green channel as a robust physical prior to locate non-local similar patches and perform principal component analysis (PCA)-based collaborative denoising. This approach achieves high-fidelity microvascular perfusion reconstruction in both few-frame temporal and single-image spatial LSCI. The proposed method suppresses the statistical noise and restores the high-fidelity perfusion map, which preserves the dynamic perfusion of microvessels.