Microscope Autofocus Based on Difference of Gaussians and Triplet Loss

Whole slide imaging (WSI) involves scanning tissue specimens on slides using a microscope and converting them into digital formats, thereby preserving comprehensive tissue information to assist clinical diagnosis. High-precision autofocus methods are crucial for ensuring the quality of WSI. However, existing autofocus methods can be affected by variations in staining methods and sample differences. This paper introduces a robust autofocus method based on Difference of Gaussians (DoG) image and triplet loss. By utilizing DoG image to emphasize image edge information that is highly correlated with focus distance, this method helps reduce the impact of staining variations. Furthermore, triplet loss enhances the network model’s sensitivity to defocus distance, effectively addressing the issue of sample shape differences. Comparative and ablation experiments conducted on publicly available datasets demonstrate that the proposed method achieves state-of-the-art performance.