Automatic deformable registration of histological slides to μCT volume data

Area of Science:

• Biomedical Imaging
• Computational Biology
• Medical Image Analysis

Background:

• 2D-3D registration of histological sections to 3D imaging datasets is a complex challenge.
• Existing methods often lack full automation and struggle with elastic distortions from tissue preparation.
• Previous automatic algorithms showed promise but had limitations in accuracy for some datasets.

Purpose of the Study:

• To introduce and evaluate two novel optimization frameworks for accurate 2D-3D registration of histology slides to micro computed tomography (μCT) volume data.
• To address the limitations of previous automatic registration methods, including handling elastic distortions.
• To improve the precision of localizing histological sections within 3D μCT volumes.

Main Methods:

• Development of two optimization frameworks based on normalized mutual information for 2D-3D image registration.
• Implementation of a rigid registration approach for histological section localization.
• Implementation of a deformable registration approach to account for elastic distortions in histological preparation.

Main Results:

• The rigid approach achieved 81% accuracy in localizing histological sections in jaw bone datasets with a median error of 8.4 μm.
• The deformable approach significantly improved registration accuracy by 33 μm in median distance error for cerebellum datasets.
• Both frameworks demonstrated enhanced accuracy in registering histology slides to volume data.

Conclusions:

• The proposed normalized mutual information-based optimization frameworks enable accurate and robust 2D-3D registration of histological sections to μCT data.
• The deformable approach effectively compensates for elastic distortions, a significant improvement over prior methods.
• These advancements facilitate more precise integration of histological findings with 3D volumetric imaging data.