SR-CACO-2: A Dataset for Confocal Fluorescence Microscopy Image Super-Resolution

TL;DR

A large scanning confocal microscopy dataset named SR-CACO-2 is introduced that is comprised of low- and high-resolution image pairs marked for three different fluorescent markers, and it allows the evaluation of performance of SISR methods on three different upscaling levels (X2, X4, X8).

Abstract

Confocal fluorescence microscopy is one of the most accessible and widely used imaging techniques for the study of biological processes at the cellular and subcellular levels. Scanning confocal microscopy allows the capture of high-quality images from thick three-dimensional (3D) samples, yet suffers from well-known limitations such as photobleaching and phototoxicity of specimens caused by intense light exposure, limiting its applications. Cellular damage can be alleviated by changing imaging parameters to reduce light exposure, often at the expense of image quality. Machine/deep learning methods for single-image super-resolution (SISR) can be applied to restore image quality by upscaling lower-resolution (LR) images to yield high-resolution images (HR). These SISR methods have been successfully applied to photo-realistic images due partly to the abundance of publicly available data. In contrast, the lack of publicly available data partly limits their application and success in scanning confocal microscopy. In this paper, we introduce a large scanning confocal microscopy dataset named SR-CACO-2 that is comprised of low- and high-resolution image pairs marked for three different fluorescent markers. It allows the evaluation of performance of SISR methods on three different upscaling levels (X2, X4, X8). SR-CACO-2 contains the human epithelial cell line Caco-2 (ATCC HTB-37), and it is composed of 2,200 unique images, captured with four resolutions and three markers, forming 9,937 image patches for SISR methods. We provide benchmarking results for 16 state-of-the-art methods of the main SISR families. Results show that these methods have limited success in producing high-resolution textures. The dataset is freely accessible under a Creative Commons license (CC BY-NC-SA 4.0). Our dataset, code and pretrained weights for SISR methods are available: https://github.com/sbelharbi/sr-caco-2.

Figures (15)

• Figure 1: Illustration of SR-CACO-2 patch content for cells CELL0, CELL1, and CELL2, and for the HR patch and its corresponding three LR patches (/2, /4, /8). The HR patch size is ${512\times 512}$, while the LR patch sizes are ${256\times 256}$, ${128\times 128}$, and ${64\times 64}$ for scales /2, /4, and /8, respectively.
• Figure 2: SISR families. The grey block indicates a predefined upsampling such as interpolation. Blue, green, and yellow indicate learnable convolution, upsampling, and downsampling modules, respectively.
• Figure 3: Methodology for capturing the SR-CACO-2 dataset follows these steps: Step 1: Lentiviral infection. mCherry-Histone H2B and GFP-tubulin lentivirus are added to a monolayer of Caco-2 epithelial cells. Infection of the monolayer with the viruses results in permanent modification of the cells to expression mCherry-Histone H2B (Magenta circles = red fluorescent chromosomes) and GFP-tubulin (Green rectangles = green fluorescent microtubules). Step 2: Growing 3D epithelial cysts. The Caco-2 monolayer is detached to form a single-cell suspension. Cell culture plates are coated in a layer of Geltrex basement membrane extract (BME), onto which the Caco-2 single-cell suspension is added. After 5 days in culture, the single cells grow to form organized multicellular 3D structures, known as cysts or spheroids. Step 3: Immunostaining. Primary antibodies target survivin (a marker of midbodies, a structure present at the end of cell division) and E-cadherin (a cell membrane marker). Secondary antibodies result in the fluorescence of these markers in either green or far red channels. Step 4: Image acquisition. Tile scans are performed with an LSM700 microscope at 4 resolution levels. All 3 channels are captured with each tile scan: Survivin (CELL0), E-cadherin or GFP-tubulin (CELL1), mCherry-Histone H2B (CELL2).
• Figure 4: Pre-processing of tiles to patches. First, the HR tile of cell CELL2 (the brightest) is binarized to allow localizing cells only (i.e. ROI). A sliding window of size ${512\times 512}$ is used to scan the entire tile with specific overlap. Only windows with enough cell mass are preserved while the rest are discarded. The same coordinates of the preserved patches are used for CELL1 and CELL2 of HR tiles. Coordinates of LR patches X2, X4, and X8 are computed by scaling down the HR patch coordinates according to the corresponding scale.
• Figure 5: Object-based analysis of cellular structures captured of all the 2,200 high-resolution images (${22\times 10\times 10}$).