Non-rigid point cloud registration for middle ear diagnostics with endoscopic optical coherence tomography

Area of Science:

• Medical Imaging
• Computational Anatomy
• Otolaryngology

Background:

• Middle ear infections are common, particularly in children, with current diagnostics relying on subjective visual otoscope examination.
• Endoscopic optical coherence tomography (OCT) offers in vivo morphological and functional data but faces interpretation challenges due to image artifacts and data incompleteness.

Purpose of the Study:

• To enhance the readability and diagnostic utility of OCT data for middle ear structures.
• To develop a computational method for faster and more accurate interpretation of in vivo OCT images.

Main Methods:

• Proposed C2P-Net, a two-stage non-rigid registration pipeline for complete-to-partial point clouds derived from ex vivo and in vivo OCT data.
• Developed a Blender3D simulation pipeline to generate synthetic middle ear shapes and extract noisy, partial point clouds for training data augmentation.

Main Results:

• C2P-Net demonstrated generalization to unseen middle ear point clouds.
• The pipeline effectively handled realistic noise and incompleteness in both synthetic and real OCT datasets.

Conclusions:

• Introduced C2P-Net for the first time to interpret in vivo noisy and partial OCT images of the middle ear.
• The developed method facilitates the diagnosis of middle ear structures using OCT, promoting its clinical application.