NuSeT: A deep learning tool for reliably separating and analyzing crowded cells

Abstract

Segmenting cell nuclei within microscopy images is a ubiquitous task in biological research and clinical applications. Unfortunately, segmenting low-contrast overlapping objects that may be tightly packed is a major bottleneck in standard deep learning-based models. We report a Nuclear Segmentation Tool (NuSeT) based on deep learning that accurately segments nuclei across multiple types of fluorescence imaging data. Using a hybrid network consisting of U-Net and Region Proposal Networks (RPN), followed by a watershed step, we have achieved superior performance in detecting and delineating nuclear boundaries in 2D and 3D images of varying complexities. By using foreground normalization and additional training on synthetic images containing non-cellular artifacts, NuSeT improves nuclear detection and reduces false positives. NuSeT addresses common challenges in nuclear segmentation such as variability in nuclear signal and shape, limited training sample size, and sample preparation artifacts. Compared to other segmentation models, NuSeT consistently fares better in generating accurate segmentation masks and assigning boundaries for touching nuclei.

Potential Relationships Between NAFLD Fibrosis Score and Graft Status in Liver Transplant Patients

“…The use of AI in histopathological assessment yields accurate, objective results that can be analyzed quickly to provide information regarding associations between histopathological features of the liver tissue and patient outcomes post-transplant. 15 Very little association was found between donor NFS and biopsy liver fibrosis in pathologist assessments (Figure 5). It is not surprising that the pathologist scoring of fibrosis was low in general, as these livers came from a database of livers accepted for donation.…”

UnMICST: Deep learning with real augmentation for robust segmentation of highly multiplexed images of human tissues

“…Subsequent applications of ImageNet make clear that machine learning models are only as strong as the underlying training and validation data. In biomedical research, segmentation methods for cell lines (Caicedo et al, 2019;Liu et al, 2020;Torr et al, 2020) have benefited from large public datasets such as Expert Visual Cell ANnotation (EVICAN) (Schwendy, Unger and Parekh, 2020), Cellpose (Stringer et al, 2020), the Broad Bioimage Benchmark Collection (BBBC) (Ljosa, Sokolnicki and Carpenter, 2012), The Cancer Genome Atlas (TCGA) (Xing et al, 2019), and Kaggle datasets (Yang et al, 2020).…”

“…A few studies have also added elastic deformations using B-splines (Ronneberger, Fischer and Brox, 2015;Raza et al, 2019;Torr et al, 2020). These methods are not unique to microscopy images, however, and only a few studies have used augmentation to address variation in the brightness and contrast of otherwise identical images (Lugagne, Lin and Dunlop, 2020) or added synthetically generated camera noise and non-cellular debris to make model training less sensitive to artefacts (Schmidt et al, 2018;Yang et al, 2020). A particularly interesting form of augmentation used by Kromp et al (2019) involved manually separating cells from the background and arranging nuclei in grids with random positions and orientations, effectively generating new training examples.…”

“…To date, common artefacts such as image saturation and defocus have been addressed computationally using image histogram modification and Gaussian blurring (Shorten and Khoshgoftaar, 2019;Yang et al, 2020) as well as deliberate defocusing of an actual microscope (Ljosa, Sokolnicki and Carpenter, 2012). The BBBC described by Ljosa et al is primarily derived from tissue culture cells or transmission and differential interference contrast (DIC) microscopy of model organisms as opposed to human tissues.…”

“…In fluorescence microscopy, nuclei are most commonly stained using intercalating dyes such as DAPI or Hoechst 33342 (Al-Kofahi et al, 2010;Schmidt et al, 2018;Berryman et al, 2019;Vuola, Akram and Kannala, 2019), SiR-DNA (Yang et al, 2020), TO-PRO (Byun et al, 2006), and hematoxylin (Al-Kofahi et al, 2010;Qi et al, 2012;Xu, Lu and Mandal, 2014;Chen et al, 2016;Xu et al, 2017;Vu et al, 2019;Lee and Jeong, 2020;Shahzad M et al, 2020). When expression of recombinant proteins is feasible (e.g., in cell lines), cells expressing fluorescent protein fusions to histones (Challen and Goodell, 2008) or spindle components is an effective means to label nuclei (Wen et al, 2018;Wang et al, 2019); this has also been done in genetically engineered mouse models (Tumbar et al, 2004) but is not relevant to analysis of human tissues.…”