ReshapeIT: Reliable Shape Interaction with Implicit Template for Anatomical Structure Reconstruction

TL;DR

ReShapeIT introduces a continuous-space framework for anatomical shape reconstruction by coupling an implicit deformation field with a shared implicit template field, enabling reliable dense correspondence and implicit generalization. A Valid Template Generation mechanism and a Template Interaction Module (TIM) enable alignment and refinement of unseen shapes, improving reconstruction accuracy over voxel-based and prior implicit methods. Quantitative results on MSD Liver, NIH Pancreas-CT, and Lung Lobe datasets show superior Chamfer Distance and Earth Mover’s Distance, with notable gains in boundary accuracy and reduced artifacts, while ablations confirm the importance of instance/hyper latent codes and regularization terms. The approach offers a plug-and-play pathway to enhance existing segmentation pipelines and supports potential extensions to multi-class templates and anatomy-aware registration, supporting more accurate and interpretable 3D medical models.

Abstract

Shape modeling of volumetric medical images is crucial for quantitative analysis and surgical planning in computer-aided diagnosis. To alleviate the burden of expert clinicians, reconstructed shapes are typically obtained from deep learning models, such as Convolutional Neural Networks (CNNs) or transformer-based architectures, followed by the marching cube algorithm. However, automatic shape reconstruction often falls short of perfection due to the limited resolution of images and the absence of shape prior constraints. To overcome these limitations, we propose the Reliable Shape Interaction with Implicit Template (ReShapeIT) network, which models anatomical structures in continuous space rather than discrete voxel grids. ReShapeIT represents an anatomical structure with an implicit template field shared within the same category, complemented by a deformation field. It ensures the implicit template field generates valid templates by strengthening the constraint of the correspondence between the instance shape and the template shape. The valid template shape can then be utilized for implicit generalization. A Template Interaction Module (TIM) is introduced to reconstruct unseen shapes by interacting the valid template shapes with the instance-wise latent codes. Experimental results on three datasets demonstrate the superiority of our approach in anatomical structure reconstruction. The Chamfer Distance/Earth Mover's Distance achieved by ReShapeIT are 0.225/0.318 on Liver, 0.125/0.067 on Pancreas, and 0.414/0.098 on Lung Lobe.

Figures (10)

• Figure 1: ReShapeIT interacts initial results with valid template shapes based on reliable dense correspondence, followed by the implicit neural representation to acquire refined reconstruction.
• Figure 2: Overview of the Reliable Shape Interaction with Implicit Template (ReShapeIT) Framework. For the instance shapes, world coordinates $\bm{p} = \{x,y,z\}$ along with corresponding latent code $\bm{\alpha}$ are fed into the implicit deform field $\mathit{Deform}$ followed by one implicit template field $\mathit{Temp}$ to construct implicit shape modeling of medical anatomical structures. The template space is sampled and assigned the latent template code $\bm{t}$, and they are also fed into the $\mathit{Deform}$ as the same way for instance shapes, aiming to build accurate correspondence between the template shape and instance shape, simultaneously achieve the reliable template shape.
• Figure 3: Template interaction module (TIM) for the implicit generalization for the reconstruction of unseen anatomical structures.
• Figure 4: Examples of the implicit reconstruction results on the Liver, Pancreas, and Right Middle Lobe. The dotted blue boxes highlight the interaction areas between the initial result and the valid template shape.
• Figure 5: Qualitative comparisons of the reconstruction error among the proposed methods with others. The dark purple indicates low reconstruction error, while bright yellow denotes high reconstruction error.