U-Net: deep learning for cell counting, detection, and morphometry

Falk, Thorsten; Mai, Dominic; Bensch, Robert; Çiçek, Özgün; Abdulkadir, Ahmed; Marrakchi, Yassine; Böhm, Anton; Deubner, Jan; Jäckel, Zoë; Seiwald, Katharina; Dovzhenko, Alexander; Tietz, Olaf; Bosco, Cristina Dal; Walsh, Seán; Saltukoglu, Deniz; Tay, Tuan Leng; Prinz, Marco; Palme, Klaus; Simons, Matias; Diester, Ilka; Brox, Thomas; Ronneberger, Olaf

doi:10.1038/s41592-018-0261-2

Cited by 1,453 publications

(1,194 citation statements)

References 13 publications

Supporting

Mentioning

1,179

Contrasting

Unclassified

Order By: Relevance

“…We did not introduce a different weighting scheme for edges between two cells, giving all boundary pixels the same weight regardless of their context (background or another cell). Also, a U-Net plugin was independently developed for ImageJ for running generic cell segmentation and quantification tasks (42). Also, we apply additional data augmentation using elastic deformations, as discussed by the authors (16).…”

Section: U-netmentioning

confidence: 99%

See 1 more Smart Citation

Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images

et al. 2019

View full text Add to dashboard Cite

Identifying nuclei is often a critical first step in analyzing microscopy images of cells and classical image processing algorithms are most commonly used for this task. Recent developments in deep learning can yield superior accuracy, but typical evaluation metrics for nucleus segmentation do not satisfactorily capture error modes that are relevant in cellular images. We present an evaluation framework to measure accuracy, types of errors, and computational efficiency; and use it to compare deep learning strategies and classical approaches. We publicly release a set of 23,165 manually annotated nuclei and source code to reproduce experiments and run the proposed evaluation methodology. Our evaluation framework shows that deep learning improves accuracy and can reduce the number of biologically relevant errors by half. © 2019 The Authors. Cytometry Part A published by Wiley Periodicals, Inc. on behalf of International Society for Advancement of Cytometry.

show abstract

Section: U-netmentioning

confidence: 99%

“…The source code of our U-Net implementation can be found in https://github.com/ carpenterlab/2019_caicedo_cytometryA, with an optional CellProfiler 3.0 plugin of this nucleus-specific model (41). Also, a U-Net plugin was independently developed for ImageJ for running generic cell segmentation and quantification tasks (42).…”

Section: U-netmentioning

confidence: 99%

Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images

et al. 2019

View full text Add to dashboard Cite

show abstract

“…This is a complicated task even for humans, and as classical computational approaches have not been sufficiently accurate, fluorescent staining with its clean nuclear signal has been preferred. However, recent breakthroughs in deep learning have led to impressive performance on image analysis tasks in general (Angermueller et al 2016;Fan and Zhou 2016) , and segmentation from cell images in particular (Van Valen et al 2016;Falk et al 2019;Christiansen et al 2018;Jones et al 2017) . This motivates a re-evaluation of whether nuclear segmentation could be achieved without a DNA stain.…”

Section: Introductionmentioning

confidence: 99%

Segmenting nuclei in brightfield images with neural networks

Fishman

Salumaa

Majoral

et al. 2019

Preprint

View full text Add to dashboard Cite

Identifying nuclei is a standard first step to analysing cells in microscopy images. The traditional approach relies on signal from a DNA stain, or fluorescent transgene expression localised to the nucleus. However, imaging techniques that do not use fluorescence can also carry useful information. Here, we demonstrate that it is possible to accurately segment nuclei directly from brightfield images using deep learning. We confirmed that three convolutional neural network architectures can be adapted for this task, with U-Net achieving the best overall performance, Mask R-CNN providing an additional benefit of instance segmentation, and DeepCell proving too slow for practical application. We found that accurate segmentation is possible using as few as 16 training images and that models trained on images from similar cell lines can extrapolate well. Acquiring data from multiple focal planes further helps distinguish nuclei in the samples. Overall, our work liberates a fluorescence channel reserved for nuclear staining, thus providing more information from the specimen, and reducing reagents and time required for preparing imaging experiments.

show abstract

“…Recently established histological methods allow now resolving the three‐dimensional (3D) structures of microglia in large mammalian brain samples (Chung et al, ; Grabow, Yoder, & Mote, ; Hama et al, ; Ke, Fujimoto, & Imai, ; Lai et al, ). A few recent corresponding computational approaches exist to analyze 3D morphologies (Falk et al, ; Heindl et al, ) but are either not designed for an unbiased high‐throughput analysis or not specialized on microglia morphologies in more complex human postmortem samples. To address this gap, we developed a computational pipeline for Microglia and Immune Cell Morphological Analysis and Classification (MIC‐MAC; https://micmac.lcsb.uni.lu/ and Supporting Information Figure S1) that captures morphological heterogeneity of microglia at single cell level in large 3D high‐resolution confocal stacks from mouse and human brain sections immunolabeled for cell‐type specific morphological markers.…”

Section: Introductionmentioning

confidence: 99%

“…Recently established histological methods allow now resolving the threedimensional (3D) structures of microglia in large mammalian brain samples (Chung et al, 2013;Grabow, Yoder, & Mote, 2000;Hama et al, 2015;Ke, Fujimoto, & Imai, 2013;Lai et al, 2018). A few recent corresponding computational approaches exist to analyze 3D morphologies (Falk et al, 2019;Heindl et al, 2018) Figure S1) that captures morphological heterogeneity of microglia at single cell level in large 3D high-resolution confocal stacks from mouse and human brain sections immunolabeled for cell-type specific morphological markers.…”

Section: Introductionmentioning

confidence: 99%

MIC‐MAC: An automated pipeline for high‐throughput characterization and classification of three‐dimensional microglia morphologies in mouse and human postmortem brain samples

Salamanca

Mechawar

Murai

et al. 2019

Glia

View full text Add to dashboard Cite

The phenotypic changes of microglia in brain diseases are particularly diverse and their role in disease progression, beneficial, or detrimental, is still elusive. High‐throughput molecular approaches such as single‐cell RNA‐sequencing can now resolve the high heterogeneity in microglia population for a specific physiological condition, however, the relation between the different microglial signatures and their surrounding brain microenvironment is barely understood. Thus, better tools to characterize the phenotypic variations of microglia in situ are needed, particularly for human brain postmortem samples analysis. To address this challenge, we developed MIC‐MAC, a Microglia and Immune Cells Morphologies Analyser and Classifier pipeline that semiautomatically segments, extracts, and classifies all microglia and immune cells labeled in large three‐dimensional (3D) confocal image stacks of mouse and human brain samples. Our imaging‐based approach enables automatic 3D‐morphology characterization and classification of thousands of individual microglia in situ and revealed species‐ and disease‐specific morphological phenotypes in mouse aging, human Alzheimer's disease, and dementia with Lewy Bodie's samples. MIC‐MAC is a precision diagnostic tool that allows a rapid, unbiased, and large‐scale analysis of microglia morphological states in mouse models and patient brain samples.

show abstract

U-Net: deep learning for cell counting, detection, and morphometry

Cited by 1,453 publications

References 13 publications

Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images

Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images

Segmenting nuclei in brightfield images with neural networks

MIC‐MAC: An automated pipeline for high‐throughput characterization and classification of three‐dimensional microglia morphologies in mouse and human postmortem brain samples

Contact Info

Product

Resources

About