Correlated Widefield-confocal Microscopy Dataset

<strong>How to cite us</strong>

Li, R., Della Maggiora, G., Andriasyan, V., Petkidis, A., Yushkevich, A., Deshpande, N., ... &amp; Yakimovich, A. (2024). Microscopy image reconstruction with physics-informed denoising diffusion probabilistic model. Communications Engineering, 3(1), 186.

<br>
@article{li2024microscopy,<br>
  title={Microscopy image reconstruction with physics-informed denoising diffusion probabilistic model},<br>
  author={Li, Rui and Della Maggiora, Gabriel and Andriasyan, Vardan and Petkidis, Anthony and Yushkevich, Artsemi and Deshpande, Nikita and Kudryashev, Mikhail and Yakimovich, Artur},<br>
  journal={Communications Engineering},<br>
  volume={3},<br>
  number={1},<br>
  pages={186},<br>
  year={2024},<br>
  publisher={Nature Publishing Group UK London}<br>
}

<strong>Download Timeout Troubleshooting</strong>

Use "-C" flag of curl in case you experience timeout of the download:

curl -C - https://rodare...tar.gz_part1\?download\=1 --output spa.tar.gz_part1

<strong>Dataset</strong>

This dataset contains a sample of 600 fluorescently labelled nuclei of cultured cells imaged using widefield fluorescence microscopy and confocal fluorescence microscopy at different focal planes.

Image preprocessing

Notably, the hardware precision of the sectioning process led to variations in the step size when shifting the focal plane between the two devices. This resulted in distinct z-dimensions between the datasets obtained from the two microscopy techniques. The confocal stacks in raw data comprised 92 focal planes, whereas the widefield stacks consisted of only 40 slices. Each focal plane image had a shape [2048, 2048, 1]. Assuming the central slice of each stack to be the in-focus, we performed z-direction registration by downsampling the confocal stacks from the central slice (46th) to match the 40 slices of the widefield stacks. Due to the instrumental limitations, a slight drift was noticeable between images. To address this, we used the phase cross-correlation algorithm [2] to compensate for the offsets on the x-y plane for the z-dimension registered image stacks. Having completed the registration and alignment along three dimensions, we then partitioned the original images into non-overlapping patches with dimensions of [128, 128, 1] in the xy plane. This partitioned dataset serves as the test dataset for validating our blind-deconvolution model, conducted without the specific Point Spread Function (PSF) parameters [3].

<strong>Files description</strong>

The Widefield-confocal Microscopy Dataset is stored in the '*.npz' format, encompassing the variables 'c_img' and 'w_img.' These handles respectively denote the confocal images and their corresponding widefield microscopy images. Both types of data undergo registration, alignment, and normalization, with values scaled to range between [0.0, 1.0]. For each category, the data has a shape of [600, 128, 128, 40], where the first dimension denotes the individual field of view and the last dimension signifies the z-dimension representing changes in the focal plane for virtual sectioning. The first dimension corresponds to the patch number, each with a patch size of [128, 128].

<strong> </strong>

<strong>Sample preparation and microscopy</strong>

A549 lung carcinoma cell line cells were seeded in 96-well imaging plates a night prior to imaging, then fixed with 4% paraformaldehyde (Sigma) and stained for DNA with Hoechst 33342 fluorescent dye (Sigma). Cell culture was maintained similarly to the procedures described in [1]. Next, stained cell nuclei were imaged using ImageXpress Confocal system (Molecular Devices) in either confocal or widefield mode employing Nikon 20X Plan Apo Lambda objective. To obtain 3D information images in both modes were acquired as Z-stacks with 0.3 µm and 0.7 µm for confocal and widefield modes respectively. Confocal z-stack was Nyquist sampled. The excitation wavelength was 405 nm and the emission was 452 nm. Using these settings, we obtained individual stacks for both modalities, with each stack covering 2048 by 2048 pixels or 699 by 699 µm.

<strong>References</strong>

Yakimovich, Artur, et al. "Plaque2. 0—a high-throughput analysis framework to score virus-cell transmission and clonal cell expansion." PloS one 10.9 (2015): e0138760.

Alink, Mark S. Oude, et al. "Lowering the SNR wall for energy detection using cross-correlation." IEEE transactions on vehicular technology 60.8 (2011): 3748-3757.

Li, Rui, et al. "Microscopy image reconstruction with physics-informed denoising diffusion probabilistic model." arXiv preprint arXiv:2306.02929 (2023).

<strong>引用方式</strong>

Li, R., Della Maggiora, G., Andriasyan, V., Petkidis, A., Yushkevich, A., Deshpande, N., ... & Yakimovich, A. (2024). 基于物理信息降噪扩散概率模型的显微图像重建. 通信工程（Communications Engineering）, 3(1), 186.

<br>
@article{li2024microscopy,<br>
&nbsp; title={基于物理信息降噪扩散概率模型的显微图像重建},<br>
&nbsp; author={Li, Rui and Della Maggiora, Gabriel and Andriasyan, Vardan and Petkidis, Anthony and Yushkevich, Artsemi and Deshpande, Nikita and Kudryashev, Mikhail and Yakimovich, Artur},<br>
&nbsp; journal={Communications Engineering},<br>
&nbsp; volume={3},<br>
&nbsp; number={1},<br>
&nbsp; pages={186},<br>
&nbsp; year={2024},<br>
&nbsp; publisher={英国伦敦自然出版集团}<br>
}

<strong>下载超时故障排查</strong>

若遇到下载超时问题，请使用curl的"-C"参数：

curl -C - https://rodare...tar.gz_part1?download=1 --output spa.tar.gz_part1

<strong>数据集说明</strong>

本数据集包含600张经荧光标记的培养细胞细胞核样本，采用宽场荧光显微镜与共聚焦荧光显微镜在不同焦平面下成像所得。

<strong>图像预处理</strong>

值得注意的是，两种设备在切换焦平面时，切片过程的硬件精度差异导致步长存在差异，进而使得两种显微技术所得数据集的z轴维度不同。原始数据中的共聚焦堆叠包含92个焦平面，而宽场堆叠仅含40个切片。每张焦平面图像的尺寸为[2048, 2048, 1]。我们以每个堆叠的中央切片作为聚焦平面，将共聚焦堆叠从中央切片（第46张）向下采样至40个切片，以匹配宽场堆叠的切片数。受仪器性能限制，图像间存在轻微漂移，为此我们使用相位互相关算法[2]对完成z轴配准后的图像堆叠在x-y平面上进行偏移补偿。完成三维配准与对齐后，我们将原始图像在xy平面上划分为尺寸为[128, 128, 1]的非重叠图像块。该划分后的数据集将作为测试集，用于验证我们的盲反卷积模型（未使用特定点扩散函数（Point Spread Function, PSF）参数）[3]。

<strong>文件说明</strong>

宽场-共聚焦显微数据集以*.npz格式存储，包含变量"c_img"和"w_img"，分别指代共聚焦图像及其对应的宽场显微图像。两类数据均经过配准、对齐与归一化处理，数值范围缩放至[0.0, 1.0]。每个类别的数据形状为[600, 128, 128, 40]，其中第一维度代表单个视场，最后一维度代表用于表征虚拟切片的焦平面变化。第一维度对应图像块编号，每个图像块尺寸为[128, 128]。

<strong>样本制备与显微成像</strong>

将A549肺癌细胞系接种于96孔成像板中，于成像前一晚培养，随后用4%多聚甲醛（Sigma）固定，并用Hoechst 33342荧光染料（Sigma）对DNA进行染色。细胞培养流程与文献[1]所述方法一致。随后使用ImageXpress共聚焦系统（Molecular Devices），采用尼康20X Plan Apo Lambda物镜，以共聚焦或宽场模式对染色后的细胞核进行成像。为获取三维信息，两种模式下均以Z堆叠的形式采集图像：共聚焦模式的步长为0.3 µm，宽场模式的步长为0.7 µm。共聚焦Z堆叠采用奈奎斯特采样。激发波长为405 nm，发射波长为452 nm。通过上述设置，我们获得了两种成像模式的独立堆叠，每个堆叠的分辨率为2048×2048像素，对应视场尺寸为699×699 µm。

<strong>参考文献</strong>

Yakimovich, Artur, et al. "Plaque2. 0——一种用于评估病毒-细胞传播与克隆细胞扩增的高通量分析框架". 《公共科学图书馆·综合》10.9 (2015): e0138760.

Alink, Mark S. Oude, et al. "降低能量检测的SNR壁垒". 《IEEE车辆技术汇刊》60.8 (2011): 3748-3757.

Li, Rui, et al. "基于物理信息降噪扩散概率模型的显微图像重建". arXiv预印本arXiv:2306.02929 (2023).

搜集汇总

背景与挑战

背景概述

该数据集包含600个荧光标记的细胞核样本，通过宽场和共聚焦显微镜在不同焦平面成像，用于显微镜图像重建研究。数据经过配准、对齐和归一化处理，以npz格式存储，包含共聚焦和宽场图像对，适用于盲去卷积模型的验证。