Colorimetry camera-based single-molecule localization microscopy (CC-STORM) employs a simple optical setup to facilitate the simultaneous imaging of two or more targets at the nanoscale, but it suffers from a high data rejection rate. A recently reported deep learning-based algorithm (called CC-DeepSTORM) reduced the data rejection rate of two-color CC-STORM from 70% to 40%, while achieving crosstalk of 1%. However, when applying this algorithm to regions with dense emitters, it faces challenges with structural artifacts and low detection rates. Here, we propose CC-DenseSTORM, featuring an attention-gated standard-convolution U-Net to eliminate structural artifacts and a dual-channel adaptive classification network for robust dye classification. Simulations demonstrate that, even at a high density of 5 emitters/µm 2 , CC-DenseSTORM improves the detection rate by 2-fold compared to CC-DeepSTORM, while maintaining the data rejection rate below 30%. In experimental imaging of multiple myeloma cells, CC-DenseSTORM achieves <1% crosstalk (matching the state of the art), thus enabling simultaneous quantification of the densities of CD38 and BCMA, offering great potential for advancing dual-target immunotherapy.
Li et al. (Thu,) studied this question.