DeepCLEM: automated registration for correlative light and electron microscopy using deep learning.

Area of Science:

• Microscopy
• Biotechnology
• Computational Biology

Background:

• Correlative Light and Electron Microscopy (CLEM) requires precise image registration between fluorescence and electron microscopy (EM) datasets.
• Current registration methods are often manual or semi-automated, relying on fluorescent stains or fiducial markers, which can be time-consuming and introduce inaccuracies.
• Differences in image contrast between fluorescence and EM modalities hinder direct automated alignment.

Purpose of the Study:

• To develop a fully automated workflow for precise image registration in CLEM.
• To overcome the limitations of manual and semi-automated registration methods in CLEM.
• To provide a robust and adaptable solution for correlating imaging data.

Main Methods:

• Introduction of DeepCLEM, a novel, fully automated workflow for CLEM registration.
• Utilizing a convolutional neural network (CNN) to predict fluorescent signals directly from EM images.
• Employing correlation-based alignment to register the predicted fluorescent signal with the experimentally measured chromatin signal.

Main Results:

• DeepCLEM achieves fully automated registration of fluorescent and EM images.
• The workflow accurately predicts fluorescent signals from EM data.
• Successful registration is demonstrated using correlation-based alignment with chromatin signals.

Conclusions:

• DeepCLEM offers a significant advancement in automating CLEM image registration.
• The Fiji plugin facilitates accessibility and integration into existing microscopy workflows.
• The methodology holds potential for adaptation to other imaging modalities and 3D datasets.