Segmentation metric misinterpretations in bioimage analysis

Quantitative evaluation of image segmentation algorithms is crucial in the field of bioimage analysis. The most common assessment scores, however, are often misinterpreted and multiple definitions coexist with the same name. Here we present the ambiguities of evaluation metrics for segmentation algorithms and show how these misinterpretations can alter leaderboards of influential competitions. We also propose guidelines for how the currently existing problems could be tackled.

SuperCUT, an unsupervised multimodal image registration with deep learning for biomedical microscopy

Numerous imaging techniques are available for observing and interrogating biological samples, and several of them can be used consecutively to enable correlative analysis of different image modalities with varying resolutions and the inclusion of structural or molecular information. Achieving accurate registration of multimodal images is essential for the correlative analysis process, but it remains a challenging computer vision task with no widely accepted solution. Moreover, supervised registration methods require annotated data produced by experts, which is limited. To address this challenge, we propose a general unsupervised pipeline for multimodal image registration using deep learning. We provide a comprehensive evaluation of the proposed pipeline versus the current state-of-the-art image registration and style transfer methods on four types of biological problems utilizing different microscopy modalities. We found that style transfer of modality domains paired with fully unsupervised training leads to comparable image registration accuracy to supervised methods and, most importantly, does not require human intervention.

Correlative Fluorescence and Raman Microscopy to Define Mitotic Stages at the Single-Cell Level: Opportunities and Limitations in the AI Era

Nowadays, morphology and molecular analyses at the single-cell level have a fundamental role in understanding biology better. These methods are utilized for cell phenotyping and in-depth studies of cellular processes, such as mitosis. Fluorescence microscopy and optical spectroscopy techniques, including Raman micro-spectroscopy, allow researchers to examine biological samples at the single-cell level in a non-destructive manner. Fluorescence microscopy can give detailed morphological information about the localization of stained molecules, while Raman microscopy can produce label-free images at the subcellular level; thus, it can reveal the spatial distribution of molecular fingerprints, even in live samples. Accordingly, the combination of correlative fluorescence and Raman microscopy (CFRM) offers a unique approach for studying cellular stages at the single-cell level. However, subcellular spectral maps are complex and challenging to interpret. Artificial intelligence (AI) may serve as a valuable solution to characterize the molecular backgrounds of phenotypes and biological processes by finding the characteristic patterns in spectral maps. The major contributions of the manuscript are: (I) it gives a comprehensive review of the literature focusing on AI techniques in Raman-based cellular phenotyping; (II) via the presentation of a case study, a new neural network-based approach is described, and the opportunities and limitations of AI, specifically deep learning, are discussed regarding the analysis of Raman spectroscopy data to classify mitotic cellular stages based on their spectral maps.

Cell lines and clearing approaches: a single-cell level 3D light-sheet fluorescence microscopy dataset of multicellular spheroids

Nowadays, three dimensional (3D) cell cultures are widely used in the biological laboratories and several optical clearing approaches have been proposed to visualize individual cells in the deepest layers of cancer multicellular spheroids. However, defining the most appropriate clearing approach for the different cell lines is an open issue due to the lack of a gold standard quantitative metric. In this article, we describe and share a single-cell resolution 3D image dataset of human carcinoma spheroids imaged using a light-sheet fluorescence microscope. The dataset contains 90 multicellular cancer spheroids derived from 3 cell lines (i.e. T-47D, 5-8F, and Huh-7D12) and cleared with 5 different protocols, precisely Clear^T, Clear^T2, CUBIC, ScaleA2, and Sucrose. To evaluate image quality and light penetration depth of the cleared 3D samples, all the spheroids have been imaged under the same experimental conditions, labelling the nuclei with the DRAQ5 stain and using a Leica SP8 Digital LightSheet microscope. The clearing quality of this dataset was annotated by 10 independent experts and thus allows microscopy users to qualitatively compare the effects of different optical clearing protocols on different cell lines. It is also an optimal testbed to quantitatively assess different computational metrics evaluating the image quality in the deepest layers of the spheroids.

A quantitative metric for the comparative evaluation of optical clearing protocols for 3D multicellular spheroids

3D multicellular spheroids quickly emerged as in vitro models because they represent the in vivo tumor environment better than standard 2D cell cultures. However, with current microscopy technologies, it is difficult to visualize individual cells in the deeper layers of 3D samples mainly because of limited light penetration and scattering. To overcome this problem several optical clearing methods have been proposed but defining the most appropriate clearing approach is an open issue due to the lack of a gold standard metric. Here, we propose a guideline for 3D light microscopy imaging to achieve single-cell resolution. The guideline includes a validation experiment focusing on five optical clearing protocols. We review and compare seven quality metrics which quantitatively characterize the imaging quality of spheroids. As a test environment, we have created and shared a large 3D dataset including approximately hundred fluorescently stained and optically cleared spheroids. Based on the results we introduce the use of a novel quality metric as a promising method to serve as a gold standard, applicable to compare optical clearing protocols, and decide on the most suitable one for a particular experiment.

Phenotypic Image Analysis Software Tools for Exploring and Understanding Big Image Data from Cell-Based Assays

Phenotypic image analysis is the task of recognizing variations in cell properties using microscopic image data. These variations, produced through a complex web of interactions between genes and the environment, may hold the key to uncover important biological phenomena or to understand the response to a drug candidate. Today, phenotypic analysis is rarely performed completely by hand. The abundance of high-dimensional image data produced by modern high-throughput microscopes necessitates computational solutions. Over the past decade, a number of software tools have been developed to address this need. They use statistical learning methods to infer relationships between a cell's phenotype and data from the image. In this review, we examine the strengths and weaknesses of non-commercial phenotypic image analysis software, cover recent developments in the field, identify challenges, and give a perspective on future possibilities.

DIC image reconstruction using an energy minimization framework to visualize optical path length distribution

Label-free microscopy techniques have numerous advantages such as low phototoxicity, simple setup and no need for fluorophores or other contrast materials. Despite their advantages, most label-free techniques cannot visualize specific cellular compartments or the location of proteins and the image formation limits quantitative evaluation. Differential interference contrast (DIC) is a qualitative microscopy technique that shows the optical path length differences within a specimen. We propose a variational framework for DIC image reconstruction. The proposed method largely outperforms state-of-the-art methods on synthetic, artificial and real tests and turns DIC microscopy into an automated high-content imaging tool. Image sets and the source code of the examined algorithms are made publicly available.