Adaptive Conditional Contrast-Agnostic Deformable Image Registration With Uncertainty Estimation.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Deformable multi-contrast image registration is complex due to non-linear intensity variations.
• Traditional methods are slow, while current learning-based approaches lack generalizability to new contrasts.
• Existing methods often fail when applied to imaging contrasts not seen during training.

Purpose of the Study:

• To develop a novel framework for adaptive, contrast-agnostic deformable image registration.
• To enable accurate registration across arbitrary imaging contrasts without prior exposure.
• To enhance the trustworthiness and reliability of deformable image registration.

Main Methods:

• Proposed an adaptive conditional contrast-agnostic deformable image registration framework (AC-CAR).
• Implemented a random convolution-based contrast augmentation scheme for generalization.
• Introduced an adaptive conditional feature modulator (ACFM) for contrast-invariant feature learning.
• Integrated a variance network for contrast-agnostic registration uncertainty estimation.

Main Results:

• AC-CAR demonstrated superior registration accuracy compared to baseline methods.
• The framework exhibited significant generalization capabilities to unseen imaging contrasts.
• The proposed ACFM and variance network improved feature consistency and registration reliability.

Conclusions:

• AC-CAR offers a robust solution for deformable multi-contrast image registration.
• The adaptive, contrast-agnostic approach overcomes limitations of existing methods.
• This framework advances the reliability and applicability of image registration in diverse medical imaging scenarios.