Self-supervised representation learning (SSRL) in computer vision relies heavily on simple image transformations such as random rotation, crops, or illumination to learn meaningful and invariant features. Despite acknowledged importance, there is a lack of comprehensive exploration of the impact of transformation choice in the literature. Our study delves into this relationship, specifically focusing on microscopy imaging with subtle cell phenotype differences. We reveal that transformation design acts as a form of either unwanted or beneficial supervision, impacting feature clustering and representation relevance. Importantly, these effects vary based on class labels in a supervised dataset. In microscopy images, transformation design significantly influences the representation, introducing imperceptible yet strong biases. We demonstrate that strategic transformation selection, based on desired feature invariance, drastically improves classification performance and representation quality, even with limited training samples.
We develop a novel method for image segmentation of 3D confocal microscopy images of emerging hematopoietic stem cells. The method is based on the theory of persistent homology and uses an optimal threshold to select the most persistent cycles in the persistence diagram. This enables the segmentation of the image’s most contrasted and representative shapes. Coupling this segmentation method with a meshing algorithm, we define a pipeline for 3D reconstruction of confocal volumes. Compared to related methods, this approach improves shape segmentation, is more ergonomic to automatize, and has fewer parameters. We apply it to the segmentation of membranes, at subcellular resolution, of cells involved in the endothelial-to-hematopoietic transition (EHT) in the zebrafish embryos.
Optical coherence tomography (OCT) and confocal microscopy are pivotal in retinal imaging, offering distinct advantages and limitations. In vivo OCT offers rapid, noninvasive imaging but can suffer from clarity issues and motion artifacts, while ex vivo confocal microscopy, providing high-resolution, cellular-detailed color images, is invasive and raises ethical concerns. To bridge the benefits of both modalities, we propose a novel framework based on unsupervised 3D CycleGAN for translating unpaired in vivo OCT to ex vivo confocal microscopy images. This marks the first attempt to exploit the inherent 3D information of OCT and translate it into the rich, detailed color domain of confocal microscopy. We also introduce a unique dataset, OCT2Confocal, comprising mouse OCT and confocal retinal images, facilitating the development of and establishing a benchmark for cross-modal image translation research. Our model has been evaluated both quantitatively and qualitatively, achieving Fréchet inception distance (FID) scores of 0.766 and Kernel Inception Distance (KID) scores as low as 0.153, and leading subjective mean opinion scores (MOS). Our model demonstrated superior image fidelity and quality with limited data over existing methods. Our approach effectively synthesizes color information from 3D confocal images, closely approximating target outcomes and suggesting enhanced potential for diagnostic and monitoring applications in ophthalmology.
Hyperplexed in-situ targeted proteomics via antibody immunodetection (i.e., >15 markers) is changing how we classify cells and tissues. Differently from other high-dimensional single-cell assays (flow cytometry, single-cell RNA sequencing), the human eye is a necessary component in multiple procedural steps: image segmentation, signal thresholding, antibody validation, and iconographic rendering. Established methods complement the human image evaluation, but may carry undisclosed biases in such a new context, therefore we re-evaluate all the steps in hyperplexed proteomics. We found that the human eye can discriminate less than 64 out of 256 gray levels and has limitations in discriminating luminance levels in conventional histology images. Furthermore, only images containing visible signals are selected and eye-guided digital thresholding separates signal from noise. BRAQUE, a hyperplexed proteomic tool, can extract, in a marker-agnostic fashion, granular information from markers which have a very low signal-to-noise ratio and therefore are not visualized by traditional visual rendering. By analyzing a public human lymph node dataset, we also found unpredicted staining results by validated antibodies, which highlight the need to upgrade the definition of antibody specificity in hyperplexed immunostaining. Spatially hyperplexed methods upgrade and supplant traditional image-based analysis of tissue immunostaining, beyond the human eye contribution.
Imaging platforms for generating highly multiplexed histological images are being continually developed and improved. Significant improvements have also been made in the accuracy of methods for automated cell segmentation and classification. However, less attention has focused on the quantification and analysis of the resulting point clouds, which describe the spatial coordinates of individual cells. We focus here on a particular spatial statistical method, the cross-pair correlation function (cross-PCF), which can identify positive and negative spatial correlation between cells across a range of length scales. However, limitations of the cross-PCF hinder its widespread application to multiplexed histology. For example, it can only consider relations between pairs of cells, and cells must be classified using discrete categorical labels (rather than labeling continuous labels such as stain intensity). In this paper, we present three extensions to the cross-PCF which address these limitations and permit more detailed analysis of multiplex images: topographical correlation maps can visualize local clustering and exclusion between cells; neighbourhood correlation functions can identify colocalization of two or more cell types; and weighted-PCFs describe spatial correlation between points with continuous (rather than discrete) labels. We apply the extended PCFs to synthetic and biological datasets in order to demonstrate the insight that they can generate.
Supervised deep learning approaches can artificially increase the resolution of microscopy images by learning a mapping between two image resolutions or modalities. However, such methods often require a large set of hard-to-get low-res/high-res image pairs and produce synthetic images with a moderate increase in resolution. Conversely, recent methods based on GAN latent search offered a drastic increase in resolution without the need of paired images. However, they offer limited reconstruction of the high-resolution image interpretable features. Here, we propose a robust super-resolution method based on regularized latent search~(RLS) that offers an actionable balance between fidelity to the ground-truth and realism of the recovered image given a distribution prior. The latter allows to split the analysis of a low-resolution image into a computational super-resolution task performed by deep learning followed by a quantification task performed by a handcrafted algorithm and based on interpretable biological features. This two-step process holds potential for various applications such as diagnostics on mobile devices, where the main aim is not to recover the high-resolution details of a specific sample but rather to obtain high-resolution images that preserve explainable and quantifiable differences between conditions.
In this paper we propose an algorithm for aligning three-dimensional objects when represented as density maps, motivated by applications in cryogenic electron microscopy. The algorithm is based on minimizing the 1-Wasserstein distance between the density maps after a rigid transformation. The induced loss function enjoys a more benign landscape than its Euclidean counterpart and Bayesian optimization is employed for computation. Numerical experiments show improved accuracy and efficiency over existing algorithms on the alignment of real protein molecules. In the context of aligning heterogeneous pairs, we illustrate a potential need for new distance functions.
Cryogenic electron tomography (cryoET) is capable of determining in situ biological structures of molecular complexes at near atomic resolution by averaging half a million subtomograms. While abundant complexes/particles are often clustered in arrays, precisely locating and seamlessly averaging such particles across many tomograms present major challenges. Here, we developed TomoNet, a software package with a modern graphical user interface to carry out the entire pipeline of cryoET and subtomogram averaging to achieve high resolution. TomoNet features built-in automatic particle picking and 3D classification functions and integrates commonly used packages to streamline high-resolution subtomogram averaging for structures in one-, two- or three-dimensional arrays. Automatic particle picking is accomplished in two complementary ways: one based on template matching and the other employing deep learning. TomoNet's hierarchical file organization and visual display facilitate efficient data management as required for large cryoET datasets. Applications of TomoNet to three types of datasets demonstrate its capability of efficient and accurate particle picking on flexible and imperfect lattices to obtain high-resolution 3D biological structures: virus-like particles, bacterial surface layers within cellular lamellae, and membranes decorated with nuclear egress protein complexes. These results demonstrate TomoNet's potential for broad applications to various cryoET projects targeting high-resolution in situ structures.
In this work, we present a pair of tools to improve the fiducial tracking and reconstruction quality of cryo-scanning transmission electron tomography (STET) datasets. We then demonstrate the effectiveness of these two tools on experimental cryo-STET data. The first tool, GoldDigger, improves the tracking of fiducials in cryo-STET by accommodating the changed appearance of highly defocussed fiducial markers. Since defocus effects are much stronger in scanning transmission electron microscopy than in conventional transmission electron microscopy, existing alignment tools do not perform well without manual intervention. The second tool, Checkers, combines image inpainting and unsupervised deep learning for denoising tomograms. Existing tools for denoising cryo-tomography often rely on paired noisy image frames, which are unavailable in cryo-STET datasets, necessitating a new approach. Finally, we make the two software tools freely available for the cryo-STET community.
We revisit the topic of common lines between projection images in single-particle cryo-electron microscopy (cryo-EM). We derive a novel low-rank constraint on a certain 2n × n matrix storing properly scaled basis vectors for the common lines between n projection images of one molecular conformation. Using this algebraic constraint and others, we give optimization algorithms to denoise common lines and recover the unknown 3D rotations associated with the images. As an application, we develop a clustering algorithm to partition a set of noisy images into homogeneous communities using common lines, in the case of discrete heterogeneity in cryo-EM. We demonstrate the methods on synthetic and experimental datasets.
This manuscript showcases the latest advancements in deepImageJ, a pivotal Fiji/ImageJ plugin for bioimage analysis in life sciences. The plugin, known for its user-friendly interface, facilitates the application of diverse pre-trained convolutional neural networks to custom data. The manuscript demonstrates several deepImageJ capabilities, particularly in deploying complex pipelines, three-dimensional (3D) image analysis, and processing large images. A key development is the integration of the Java Deep Learning Library, expanding deepImageJ’s compatibility with various deep learning (DL) frameworks, including TensorFlow, PyTorch, and ONNX. This allows for running multiple engines within a single Fiji/ImageJ instance, streamlining complex bioimage analysis workflows. The manuscript details three case studies to demonstrate these capabilities. The first case study explores integrated image-to-image translation followed by nuclei segmentation. The second case study focuses on 3D nuclei segmentation. The third case study showcases large image volume segmentation and compatibility with the BioImage Model Zoo. These use cases underscore deepImageJ’s versatility and power to make advanced DLmore accessible and efficient for bioimage analysis. The new developments within deepImageJ seek to provide a more flexible and enriched user-friendly framework to enable next-generation image processing in life science.
Cryo-electron microscopy (cryo-EM) is an imaging technique that allows the visualization of proteins and macromolecular complexes at near-atomic resolution. The low electron doses used to prevent radiation damage to the biological samples result in images where the power of noise is 100 times stronger than that of the signal. Accurate identification of proteins from these low signal-to-noise ratio (SNR) images is a critical task, as the detected positions serve as inputs for the downstream 3D structure determination process. Current methods either fail to identify all true positives or result in many false positives, especially when analyzing images from smaller-sized proteins that exhibit extremely low contrast, or require manual labeling that can take days to complete. Acknowledging the fact that accurate protein identification is dependent upon the visual interpretability of micrographs, we propose a framework that can perform denoising and detection in a joint manner and enable particle localization under extremely low SNR conditions using self-supervised denoising and particle identification from sparsely annotated data. We validate our approach on three challenging single-particle cryo-EM datasets and projection images from one cryo-electron tomography dataset with extremely low SNR, showing that it outperforms existing state-of-the-art methods used for cryo-EM image analysis by a significant margin. We also evaluate the performance of our algorithm under decreasing SNR conditions and show that our method is more robust to noise than competing methods.
Single-particle cryogenic electron microscopy (cryo-EM) is an imaging technique capable of recovering the high-resolution three-dimensional (3D) structure of biological macromolecules from many noisy and randomly oriented projection images. One notable approach to 3D reconstruction, known as Kam's method, relies on the moments of the two-dimensional (2D) images. Inspired by Kam's method, we introduce a rotationally invariant metric between two molecular structures, which does not require 3D alignment. Further, we introduce a metric between a stack of projection images and a molecular structure, which is invariant to rotations and reflections and does not require performing 3D reconstruction. Additionally, the latter metric does not assume a uniform distribution of viewing angles. We demonstrate the uses of the new metrics on synthetic and experimental datasets, highlighting their ability to measure structural similarity.
We propose a neural network architecture and a training procedure to estimate blurring operators and deblur images from a single degraded image. Our key assumption is that the forward operators can be parameterized by a low-dimensional vector. The models we consider include a description of the point spread function with Zernike polynomials in the pupil plane or product-convolution expansions, which incorporate space-varying operators. Numerical experiments show that the proposed method can accurately and robustly recover the blur parameters even for large noise levels. For a convolution model, the average signal-to-noise ratio of the recovered point spread function ranges from 13 dB in the noiseless regime to 8 dB in the high-noise regime. In comparison, the tested alternatives yield negative values. This operator estimate can then be used as an input for an unrolled neural network to deblur the image. Quantitative experiments on synthetic data demonstrate that this method outperforms other commonly used methods both perceptually and in terms of SSIM. The algorithm can process a 512 512 image under a second on a consumer graphics card and does not require any human interaction once the operator parameterization has been set up.1.
With the fast development of modern microscopes and bioimaging techniques, an unprecedentedly large amount of imaging data are being generated, stored, analyzed, and even shared through networks. The size of the data poses great challenges for current data infrastructure. One common way to reduce the data size is by image compression. This present study analyzes classic and deep learning based image compression methods, and their impact on deep learning based image processing models. Deep learning based label-free prediction models (i.e., predicting fluorescent images from bright field images) are used as an example application for comparison and analysis. Effective image compression methods could help reduce the data size significantly without losing necessary information, and therefore reduce the burden on data management infrastructure and permit fast transmission through the network for data sharing or cloud computing. To compress images in such a wanted way, multiple classical lossy image compression techniques are compared to several AI-based compression models provided by and trained with the CompressAI toolbox using python. These different compression techniques are compared in compression ratio, multiple image similarity measures and, most importantly, the prediction accuracy from label-free models on compressed images. We found that AI-based compression techniques largely outperform the classic ones and will minimally affect the downstream label-free task in 2D cases. In the end, we hope the present study could shed light on the potential of deep learning based image compression and the impact of image compression on downstream deep learning based image analysis models.
Abstract This paper presents a deep-learning-based workflow to detect synapses and predict their neurotransmitter type in the primitive chordate Ciona intestinalis (Ciona) electron microscopic (EM) images. Identifying synapses from EM images to build a full map of connections between neurons is a labor-intensive process and requires significant domain expertise. Automation of synapse classification would hasten the generation and analysis of connectomes. Furthermore, inferences concerning neuron type and function from synapse features are in many cases difficult to make. Finding the connection between synapse structure and function is an important step in fully understanding a connectome. Class Activation Maps derived from the convolutional neural network provide insights on important features of synapses based on cell type and function. The main contribution of this work is in the differentiation of synapses by neurotransmitter type through the structural information in their EM images. This enables the prediction of neurotransmitter types for neurons in Ciona, which were previously unknown. The prediction model with code is available on GitHub.
Abstract In vivo fluorescence microscopy is a powerful tool to image the beating heart in its early development stages. A high acquisition frame rate is necessary to study its fast contractions, but the limited fluorescence intensity requires sensitive cameras that are often too slow. Moreover, the problem is even more complex when imaging distinct tissues in the same sample using different fluorophores. We present Paired Alternating AcQuisitions, a method to image cyclic processes in multiple channels, which requires only a single (possibly slow) camera. We generate variable temporal illumination patterns in each frame, alternating between channel-specific illuminations (fluorescence) in odd frames and a motion-encoding brightfield pattern as a common reference in even frames. Starting from the image pairs, we find the position of each reference frame in the cardiac cycle through a combination of image-based sorting and regularized curve fitting. Thanks to these estimated reference positions, we assemble multichannel videos whose frame rate is virtually increased. We characterize our method on synthetic and experimental images collected in zebrafish embryos, showing quantitative and visual improvements in the reconstructed videos over existing nongated sorting-based alternatives. Using a 15 Hz camera, we showcase a reconstructed video containing two fluorescence channels at 100 fps.
High-resolution and multiplexed imaging techniques are giving us an increasingly detailed observation of a biological system. However, sharing, exploring, and customizing the visualization of large multidimensional images can be a challenge. Here, we introduce Samui, a performant and interactive image visualization tool that runs completely in the web browser. Samui is specifically designed for fast image visualization and annotation and enables users to browse through large images and their selected features within seconds of receiving a link. We demonstrate the broad utility of Samui with images generated with two platforms: Vizgen MERFISH and 10x Genomics Visium Spatial Gene Expression. Samui along with example datasets is available at https://samuibrowser.com.
Large-scale multiplex tissue analysis aims to understand processes such as development and tumor formation by studying the occurrence and interaction of cells in local environments in e.g. tissue samples from patient cohorts. A typical procedure in the analysis is to delineate individual cells, classify them into cell types, and analyze their spatial relationships. All steps come with a number of challenges, and to address them and identify the bottlenecks of the analysis, it is necessary to include quality control tools in the analysis workflow. This makes it possible to optimize the steps and adjust settings in order to get better and more precise results. Additionally, the development of automated approaches for tissue analysis requires visual verification to reduce skepticism with regard to the accuracy of the results. Quality control tools could be used to build users’ trust in automated approaches. In this paper, we present three plugins for visualization and quality control in large-scale multiplex tissue analysis of microscopy images. The first plugin focuses on the quality of cell staining, the second one was made for interactive evaluation and comparison of different cell classification results, and the third one serves for reviewing interactions of different cell types.
High-content screening (HCS) provides an excellent tool to understand the mechanism of action of drugs on disease-relevant model systems. Careful selection of fluorescent labels (FLs) is crucial for successful HCS assay development. HCS assays typically comprise (a) FLs containing biological information of interest, and (b) additional structural FLs enabling instance segmentation for downstream analysis. However, the limited number of available fluorescence microscopy imaging channels restricts the degree to which these FLs can be experimentally multiplexed. In this article, we present a segmentation workflow that overcomes the dependency on structural FLs for image segmentation, typically freeing two fluorescence microscopy channels for biologically relevant FLs. It consists in extracting structural information encoded within readouts that are primarily biological, by fine-tuning pre-trained state-of-the-art generalist cell segmentation models for different combinations of individual FLs, and aggregating the respective segmentation results together. Using annotated datasets that we provide, we confirm our methodology offers improvements in performance and robustness across several segmentation aggregation strategies and image acquisition methods, over different cell lines and various FLs. It thus enables the biological information content of HCS assays to be maximized without compromising the robustness and accuracy of computational single-cell profiling.
