Videos of Embryonic Wound Healing in Drosophila and Wound Segmentation

Abstract

We collect a dataset of time-lapse sequences of <em>Drosophila</em> embryos healing after a laser-induced wound. Altogether we acquire 61 sequences, which we split into train, validation, and test sets of 44, 12, and 5 sequences, respectively. <br> We use embryos expressing a GFP tagged myosin II (sqhGFP) imaged at stage 17. The embryos are collected after aging for 16 hours at 18°C. They are mounted on a coverslip covered with heptane glue for conventional confocal microscopy and<br> by mineral oil to prevent drying. Laser ablation is performed with a pulsed laser. The time-lapse sequences are acquired <em>in vivo </em>using an Olympus IXplore SpinSR10 at room temperature. The samples are illuminated by laser with \(488\,\textrm{nm} \) wavelength using a \(60\, \times \, 1.42 \) NA oil immersion objective. The individual frames are taken once every minute. Each sequence captures the entire closure of the wound, resulting in an average length of 120 frames. Each frame has \(1152\times1152\) pixels and captures the whole embryo with \(4.6\, \mu \textrm{m/pixel}\) image resolution. A z-stack of the embryo is imaged at each time step with a z-step size of \(1\, \mu \textrm{m}\). <br> The time-resolved z-stacks are max-projected along the z-axis, and registered by SIFT algorithm in Fiji. The wounds are segmented by a custom stable U-Net-like architecture and manually refined in napari. As each frame includes the entire embryo, it is cropped to a region of \(256\times 256\) pixels containing the wound and downsampled to \(128\times 128\) by bicubic interpolation.<br> <br> The processed timelapse sequences are in the folder Video of the Dataset.zip, and their skeletonized segmentations in the folder Segmentations.