Expected affine: A registration method for damaged section in serial sections electron microscopy.

Area of Science:

• Neuroscience
• Microscopy
• Computational Biology

Background:

• Accurate 3D reconstruction of biological tissues from serial section electron microscope (ssEM) images requires precise registration.
• Damage to serial sections during preparation poses a significant challenge for conventional registration methods, hindering neuron tracking and reconstruction.

Purpose of the Study:

• To develop a general registration method capable of handling damaged serial sections for ssEM image analysis.
• To improve the accuracy and feasibility of 3D reconstruction of biological tissues.

Main Methods:

• Key point extraction and description followed by mutual nearest neighbor matching.
• K-means and Random Sample Consensus (RANSAC) for clustering key points and approximating local affine matrices.
• K-nearest neighbor (KNN) for probability density estimation and calculation of expected affine matrices, utilizing path distance instead of Euclidean distance.

Main Results:

• The proposed method successfully registers damaged serial sections, overcoming limitations of common registration techniques.
• Experimental results on real ssEM images demonstrate the method's effectiveness in handling section damage.
• The approach facilitates more reliable 3D reconstruction of biological tissues.

Conclusions:

• This novel registration method provides a robust solution for reconstructing 3D biological tissues from damaged ssEM images.
• The technique enhances the accuracy of neuron tracking and overall 3D reconstruction pipelines.
• The developed method contributes significantly to the field of electron microscopy image analysis.