Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

POLAFFINI: Efficient feature-based polyaffine initialization for improved non-linear image registration

Antoine Legouhy, Ross Callaghan, Hojjat Azadbakht, Hui Zhang

TL;DR

The paper introduces a feature-based polyaffine initialization for non-linear image registration that leverages deep-learning segmentation to derive anatomically grounded local affine matchings. These local transformations are fused into a dense, diffeomorphic map via the log-Euclidean polyaffine (LEPT) framework, enabling faster and more robust pre-alignment than traditional affine methods. Across three diverse datasets and both traditional and DL-based registration pipelines, the polyaffine initialization yields significantly higher Dice overlap in anatomical regions—particularly the cortex—without compromising topology. The approach demonstrates strong practical impact, offering an efficient, diffeomorphic, and segmentation-informed starting point for modern image registration workflows, with publicly available code for replication.

Abstract

This paper presents an efficient feature-based approach to initialize non-linear image registration. Today, nonlinear image registration is dominated by methods relying on intensity-based similarity measures. A good estimate of the initial transformation is essential, both for traditional iterative algorithms and for recent one-shot deep learning (DL)-based alternatives. The established approach to estimate this starting point is to perform affine registration, but this may be insufficient due to its parsimonious, global, and non-bending nature. We propose an improved initialization method that takes advantage of recent advances in DL-based segmentation techniques able to instantly estimate fine-grained regional delineations with state-of-the-art accuracies. Those segmentations are used to produce local, anatomically grounded, feature-based affine matchings using iteration-free closed-form expressions. Estimated local affine transformations are then fused, with the log-Euclidean polyaffine framework, into an overall dense diffeomorphic transformation. We show that, compared to its affine counterpart, the proposed initialization leads to significantly better alignment for both traditional and DL-based non-linear registration algorithms. The proposed approach is also more robust and significantly faster than commonly used affine registration algorithms such as FSL FLIRT.

POLAFFINI: Efficient feature-based polyaffine initialization for improved non-linear image registration

TL;DR

The paper introduces a feature-based polyaffine initialization for non-linear image registration that leverages deep-learning segmentation to derive anatomically grounded local affine matchings. These local transformations are fused into a dense, diffeomorphic map via the log-Euclidean polyaffine (LEPT) framework, enabling faster and more robust pre-alignment than traditional affine methods. Across three diverse datasets and both traditional and DL-based registration pipelines, the polyaffine initialization yields significantly higher Dice overlap in anatomical regions—particularly the cortex—without compromising topology. The approach demonstrates strong practical impact, offering an efficient, diffeomorphic, and segmentation-informed starting point for modern image registration workflows, with publicly available code for replication.

Abstract

This paper presents an efficient feature-based approach to initialize non-linear image registration. Today, nonlinear image registration is dominated by methods relying on intensity-based similarity measures. A good estimate of the initial transformation is essential, both for traditional iterative algorithms and for recent one-shot deep learning (DL)-based alternatives. The established approach to estimate this starting point is to perform affine registration, but this may be insufficient due to its parsimonious, global, and non-bending nature. We propose an improved initialization method that takes advantage of recent advances in DL-based segmentation techniques able to instantly estimate fine-grained regional delineations with state-of-the-art accuracies. Those segmentations are used to produce local, anatomically grounded, feature-based affine matchings using iteration-free closed-form expressions. Estimated local affine transformations are then fused, with the log-Euclidean polyaffine framework, into an overall dense diffeomorphic transformation. We show that, compared to its affine counterpart, the proposed initialization leads to significantly better alignment for both traditional and DL-based non-linear registration algorithms. The proposed approach is also more robust and significantly faster than commonly used affine registration algorithms such as FSL FLIRT.
Paper Structure (14 sections, 3 figures, 2 tables)

This paper contains 14 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Experimental design
  4. Data
  5. Segmentation
  6. Registration
  7. Affine initialization:
  8. Polyaffine initialization:
  9. Traditional non-linear registration:
  10. Deep-learning non-linear registration:
  11. Evaluation
  12. Results
  13. Discussion
  14. Conclusion

Figures (3)

  • Figure 1: Illustration of several steps of the polyaffine estimation. a) Two sets of paired points (plain disc: moving, circle: reference). b) Delaunay triangulation performed on the reference set (blue) and pattern reproduced on the moving one (red). c) Local affine regression between two homologous neighborhoods. d) Set of estimated local affine transformations. e) Color grading representing all weight maps combined. f) Overall polyaffine transformation. Black rectangles represent the frame of the reference image.
  • Figure 2: Dice scores after non-linear registration initialized with affine and proposed polyaffine transformation for ADNI (red, HC/MCI/AD left to right), IXI (green) and UK Biobank (blue) subjects. Thick lines represent medians across all datasets.
  • Figure 3: Evolution of image similarity (LCC) and segmentation (average Dice on all regions) losses during training of deep-learning models with affine (blue) and proposed polyaffine (red) initialization for training (light) and validation (dark) samples.

Theorems & Definitions (5)

  • remark thmcounterremark
  • remark thmcounterremark
  • remark thmcounterremark
  • remark thmcounterremark
  • remark thmcounterremark