Multi-objective dual simplex-mesh based deformable image registration for 3D medical images - proof of concept

Andreadis, Georgios; Bosman, Peter A. N.; Alderliesten, Tanja

doi:10.1117/12.2605498

Cited by 2 publications

(6 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Dual-dynamic mesh transformation model The dual-dynamic mesh transformation model considered in this work uses triangular meshes (in 2D) or tetrahedral meshes (in 3D) on both images to express transformations. 15,16 Unlike in the B-spline model, the points of both the source and the target mesh are movable, allowing the model to naturally capture (dis)appearing structures and large deformations. On 3D images, each tetrahedron in the target mesh (see Figure 1c) has a corresponding tetrahedron in the source mesh, defining the spatial correspondence of the two image regions that the tetrahedra cover.…”

Section: Transformation Modelsmentioning

confidence: 99%

“…For the mesh transformation model, we use the existing MOREA registration approach, 8 which already uses MO-RV-GOMEA to optimize a custom initialized dual-dynamic mesh with hardware-acceleration on the Graphics Processing Unit (GPU). In this approach, all coordinates of mesh points connected by an edge are modified together.…”

Section: Optimization With Mo-rv-gomeamentioning

confidence: 99%

“…For each tetrahedron, it also includes 4 spoke edges that better capture flattening motion, giving a total of 10 edges. 16 Using Hooke's law, 18 the objective is defined as follows:…”

Section: Optimization Objectivesmentioning

confidence: 99%

“…While B-spline models tend to be optimized using gradient-descent methods, 6,7 mesh models are often optimized using FEM solvers such as preconditioned conjugate gradient algorithms, for forward simulation together with some other algorithm for boundary condition optimization, 4 or evolutionary algorithms (EAs), for direct optimization. 8 Gradient-descent methods are known to converge quickly, but also to be prone to converge to local optima, which can prevent these methods from fully leveraging the potential of the transformation model. EAs can be slower to converge, but are often less likely to get stuck in local optima, especially when there are many parameters to optimize.…”

Section: Introductionmentioning

confidence: 99%

“…The Multi-Objective Real-Valued Gene-pool Optimal Mixing Evolutionary Algorithm (MO-RV-GOMEA) has proven effective in optimizing a mesh transformation model within the MOREA registration approach. 8 Since using mesh models presents a complex, discontinuous optimization landscape, and the DIR problem has been shown to be inherently multi-objective, 13 the MO-RV-GOMEA optimization method is a suitable candidate to optimize both transformation models. To this end, we introduce a novel, multi-objective B-spline DIR approach optimized with MO-RV-GOMEA.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A tournament of transformation models: B-Spline-based vs. mesh-based multi-objective deformable image registration

Andreadis,

Mulder,

Bouter

et al. 2024

Medical Imaging 2024: Image Processing

View full text Add to dashboard Cite

The transformation model is an essential component of any deformable image registration approach. It provides a representation of physical deformations between images, thereby defining the range and realism of registrations that can be found. Two types of transformation models have emerged as popular choices: B-spline models and mesh models. Although both models have been investigated in detail, a direct comparison has not yet been made, since the models are optimized using very different optimization methods in practice. B-spline models are predominantly optimized using gradient-descent methods, while mesh models are typically optimized using finite-element method solvers or evolutionary algorithms. Multi-objective optimization methods, which aim to find a diverse set of high-quality trade-off registrations, are increasingly acknowledged to be important in deformable image registration. Since these methods search for a diverse set of registrations, they can provide a more complete picture of the capabilities of different transformation models, making them suitable for a comparison of models. In this work, we conduct the first direct comparison between B-spline and mesh transformation models, by optimizing both models with the same state-of-the-art multi-objective optimization method, the Multi-Objective Real-Valued Gene-pool Optimal Mixing Evolutionary Algorithm (MO-RV-GOMEA). The combination with B-spline transformation models, moreover, is novel. We experimentally compare both models on two different registration problems that are both based on pelvic CT scans of cervical cancer patients, featuring large deformations. Our results, on three cervical cancer patients, indicate that the choice of transformation model can have a profound impact on the diversity and quality of achieved registration outcomes.

show abstract

Section: Transformation Modelsmentioning

confidence: 99%

Section: Optimization With Mo-rv-gomeamentioning

confidence: 99%

“…For each tetrahedron, it also includes 4 spoke edges that better capture flattening motion, giving a total of 10 edges. 16 Using Hooke's law, 18 the objective is defined as follows:…”

Section: Optimization Objectivesmentioning

confidence: 99%