DeepProjection: specific and robust projection of curved 2D tissue sheets from 3D microscopy using deep learning.

Area of Science:

• Developmental Biology
• Bioimaging
• Computational Biology

Background:

• Extracting image data from curved tissues in 3D volumetric imaging is challenging for quantitative embryogenesis research.
• Existing methods struggle with background noise and artifacts from out-of-plane fluorescence.

Purpose of the Study:

• To develop a robust and efficient algorithm for extracting 2D images from 3D volumetric data of curved tissues.
• To address limitations in current quantitative embryogenesis imaging techniques.

Main Methods:

• Introduced DeepProjection (DP), a trainable deep learning-based projection algorithm.
• DP classifies 3D stack content and predicts binary masks to isolate target tissues.
• Masked data is projected to create background-free 2D images with preserved fluorescence intensity.

Main Results:

• DP successfully generates background-free 2D images from 3D volumetric data.
• The algorithm effectively masks out-of-plane fluorescent artifacts.
• Applied DP to analyze tissue sheet dynamics in Drosophila and Danio embryos.

Conclusions:

• DeepProjection provides a powerful new tool for quantitative analysis in developmental biology.
• The method enables accurate tracking of dynamic tissue movements and cell fate studies.
• DP is available as a documented Python package for broader research application.