Fig. 2: IMC-Denoise enables background noise removal and enhances downstream analysis of the human bone marrow IMC dataset.

a Examples of DIMR and DeepSNiF-processed IMC images labeled with CD34 and Collagen III. b Visual inspection of background removal results of DIMR and DeepSNiF-processed images, in which DIMR_thresh and DeepSNiF_thresh are binarized with the optimal thresholds (Supplementary Figs. 25 and 26), DIMR_Ilastik and DeepSNiF_Ilastik are segmented by the Ilastik software package, and MAUI results are the DIMR images processed by the MAUI software package (Supplementary Fig. 27), respectively. Manual annotated images are served as ground truths. c After DeepSNiF denoising, the background removal accuracy improves significantly in terms of F1 score, for both CD34 and Collagen III-labeled images (n = 15 independent images for CD34 and n = 12 independent images for Collagen III). Notably, DeepSNiF_Ilastik achieves the highest accuracy, while DeepSNiF_thresh performs better than all the background removal results from DIMR images. Box center indicates median, box edges 25th and 75th percentile, and whiskers minimum and maximum percentile. P values were calculated through two-sided Wilcoxon matched-paired test (**P < 0.01, ***P < 0.001, and ****P < 0.0001). d Visual inspection of DeepSNiF and DIMR_Ilastik-based denoising results on different markers-labeled IMC images. e–h Evaluations of denoising algorithms with manual gating strategies on single-cell data. The numbers in these panels are the cell percentages of the corresponding ranges. DIMR slightly enhances the single-cell analysis over raw data, while DeepSNiF further enhances the DIMR results and overall performs better than semi-automated DIMR_Ilastik-processing. Scale bar: a Top: 50 μm, bottom: 35 μm. d 107 μm.