IMC-Denoise: a content aware denoising pipeline to enhance Imaging Mass Cytometry

Article https://doi.org/10.1038/s41467-023-37123-6 IMC-Denoise: a content aware denoising pipeline to enhance Imaging Mass Cytometry Received: 15 September 2022 Accepted: 2 March 2023 Peng Lu 1,2,3,11, Karolyn A. Oetjen 4,11, Diane E. Bender 5, Marianna B. Ruzinova6, Daniel A. C. Fisher4, Kevin G. Shim4, Russell K. Pachynski4, W. Nathaniel Brennen 7,8, Stephen T. Oh 4,5,6, Daniel C. Link 4,6 & Daniel L. J. Thorek 1,2,3,9,10 Imaging Mass Cytometry (IMC) is an emerging multiplexed imaging technology for analyzing complex microenvironments using more than 40 molecularly-specific channels. However, this modality has unique data processing requirements, particularly for patient tissue specimens where signalto-noise ratios for markers can be low, despite optimization, and pixel intensity artifacts can deteriorate image quality and downstream analysis. Here we demonstrate an automated content-aware pipeline, IMC-Denoise, to restore IMC images deploying a differential intensity map-based restoration (DIMR) algorithm for removing hot pixels and a self-supervised deep learning algorithm for shot noise image filtering (DeepSNiF). IMC-Denoise outperforms existing methods for adaptive hot pixel and background noise removal, with significant image quality improvement in modeled data and datasets from multiple pathologies. This includes in technically challenging human bone marrow; we achieve noise level reduction of 87% for a 5.6-fold higher contrastto-noise ratio, and more accurate background noise removal with approximately 2 × improved F1 score. Our approach enhances manual gating and automated phenotyping with cell-scale downstream analyses. Verified by manual annotations, spatial and density analysis for targeted cell groups reveal subtle but significant differences of cell populations in diseased bone marrow. We anticipate that IMC-Denoise will provide similar benefits across mass cytometric applications to more deeply characterize complex tissue microenvironments. 1234567890():,; 1234567890():,; Check for updates Disease states are the result of a complex interplay of many different cell types interacting in close proximity in the context of often heterogeneous tissues. Traditional approaches to study these features at the tissue scale have been limited in the number of specific markers that can be acquired to robustly resolve distinct cell types. Flow cytometry, perhaps the most widely used technique to study cell populations and states in this milieu, requires single-cell disaggregation of the tissue resulting in complete loss of spatial context1,2. Highly multiplexed imaging provides a means to assess these events at A full list of affiliations appears at the end of the paper. Nature Communications | (2023)14:1601 cellular resolution in situ, with extensive protocol development in progress3, including tissue-based cyclic immunofluorescence (t-CyCIF)4, co-detection by indexing (CODEX)5, Multiplexed Ion Beam Imaging (MIBI)6,7, and Imaging Mass Cytometry (IMC)8. In IMC, tissue sections are stained with a panel of metal-conjugated antibodies, and data is acquired by UV-laser raster ablation of the section in 1-micron pixels for cytometry by time-of-flight (CyTOF) mass analyzer. This novel imaging technology allows for the detection of more than 40 antigens simultaneously to facilitate single-cell, spatially resolved, e-mail: 1 Article highly multiplexed analysis of solid tissues. This provides essential information on the distribution of transcripts, proteins, and protein modifications within single cells, microenvironments, and entire tissues8–17. The pixel data is processed into an image, thereby allowing the visualization of phenotypes and incorporation of spatial information in subsequent analyses. These properties make it a unique tool for the evaluation of complex biological systems. Despite the wide applications in pre- and clinical research using this state-of-the-art multiplexed imaging technique, there exist specific technical noise sources in IMC, which include hot pixels, channel spillover and shot noise8–10,15,18,19. Hot pixels are concentrated areas of high counts which are uncorrelated with any biological structures. Putatively, these can result from deposition of metal-stained antibody aggregates. In IMC images, single hot pixels are the most common outliers, and small hot clusters with several consecutive pixels may also exist. Channel spillover refers to scenarios where the signal of a source channel contaminates a target channel or is correlated with such contamination. The spillover in IMC can occur from a variety of reasons, such as instrument properties (abundance sensitivity), isotopic impurities and oxidation. Finally, shot noise exists because of ion counting imaging processes, which are pixel-independent, signaldependent and usually modeled as a Poisson process. Additionally, noise levels are related to multiple other factors, including variations in conjugated metal isotopes, antibody concentration and arrangement. Together these noise sources appreciably deteriorate image quality and distort downstream analyses of IMC data. Differing from traditional fluorescence-based imaging modalities, there are low background features and no read-out noises from imaging sensors in IMC. A number of studies have attempted to address the unique imaging data features of IMC. Hot pixels can be corrected by thresholding methods10,14,15,20; however, due to the differences between marker channels and tissues, a threshold needs to be pre-set carefully. An inappropriate threshold may lead to unsatisfactory results. Postacquisition methods10,19 and a bead-based compensation workflow18 have been proposed to correct the channel spillover phenomenon. However, spillover correction may not be necessary if the marker panel employed is well-designed and titrated; and the intensity of channel-overlapping signal is often weak18. Therefore, spillover can be neglected when using low concentrations of staining antibodies, which however further lowers signal-to-noise ratio (SNR). To account for the impact of shot noise, MAUI7,19 and a semi-automated Ilastik-based method21 have been used for background noise removal. These approaches require finely tuned parameters or manually annotated background regions, requiring preprocessing expertize. In tissues with low marker signals, highly intermixed cell populations, or difficult immunostaining defining thresholds can be time consuming with high inter-user subjectivity, which may still result in poor image quality that complicates further analyses. In the present work we develop and apply IMC-Denoise, a content aware denoising pipeline to enhance IMC images through an automated process. To account for the two major noise sources in this modality, hot pixels and shot noise, IMC-Denoise invokes novel algorithms for differential intensity map-based restoration (DIMR) and selfsupervised deep learning-based shot noise image filtering (DeepSNiF). We (...truncated)