3D Deformation Model for Non-Rigid Image Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for aligning 3D images of organs, such as lungs, with nonlinear deformations due to respiration, face challenges in achieving accurate alignment when linear transformations like affine transformations are used as rough alignment, leading to inaccuracies in subsequent local alignment.

Innovation Solution

An image alignment device and method that utilizes a 3D deformation model representing nonlinear changes in the target organ, allowing for precise estimation of pixel positions and non-rigid alignment between 3D images acquired at different respiratory phases, using B-spline transformation functions and quasi-Newton methods to accurately match image features.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If linear transformation (affine transformation) is used for rough alignment, then the alignment process can be performed efficiently, but the alignment accuracy deteriorates in regions with nonlinear deformation

Engineering Contradiction:
Improvealignment processing efficiencyVSAvoidalignment accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent divides the alignment process into two distinct stages: rough alignment using linear transformation for efficient global positioning, and close local alignment using non-rigid registration for accurate local deformation correction. This segmentation allows each stage to optimize for its specific purpose without compromising the other.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs rough alignment as a preliminary step before close local alignment. By first applying linear transformation to achieve approximate positioning, the patent prepares the images for the subsequent detailed non-rigid registration, ensuring that the close local alignment operates on pre-positioned images which improves overall efficiency.

Inventive Principle:
Principle #10Preliminary action

2Device complexity

If linear transformation is used for rough alignment, then the computational complexity is reduced, but the alignment precision deteriorates due to inability to capture nonlinear deformation

Engineering Contradiction:
Improvecomputational complexityVSAvoidalignment precision
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The patent segments the alignment task into coarse global alignment (low complexity) and fine local alignment (higher complexity). The linear transformation handles the computationally simpler global positioning, while the B-spline based non-rigid registration handles the computationally intensive local deformation correction only where needed.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different transformation qualities to different regions: linear transformation for the global structure and B-spline non-rigid transformation for local regions requiring detailed deformation modeling. This local quality approach ensures high precision where needed while maintaining overall computational efficiency.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS9965858B2Image alignment device, method, and program, and method for generating 3-D deformation model
Publication Date: 2018.05.08 FUJIFILM CORP

AI summary

A first 3D image and a second 3D image imaged a target organ in different phases of respiration are acquired. A 3D deformation model of the target organ which is stored in advance and represents nonlinear 3D deformation of the target organ due to respiration, and which has been generated based on information about movement of the target organ due to respiration of plural patients, is read. The positions of pixels on the second 3D image representing the same positions on the target organ as plural sampled pixels in a target organ region on the first 3D image are estimated using displacement due to changes in phase of points on the 3D deformation model corresponding to the positions on the target organ represented by the pixels. Non-rigid alignment is performed between the first 3D image and the second 3D image using the estimated positions of the pixels.