CAFusion: Controllable Anatomical Synthesis of Perirectal Lymph Nodes via SDF-guided Diffusion

TL;DR

CAFusion addresses the scarcity and bias of annotated medical imaging data by introducing an SDF-guided diffusion framework for controllable 3D synthesis of perirectal lymph nodes. It decouples morphology and texture through Anatomical Guidance (AG) and Textural Guidance (TG), plus a masked repaint step to ensure background coherence, enabling precise shape control and realistic textures. Experimental results show substantial improvements in segmentation performance when trained with synthetic data and demonstrate radiologists’ difficulty in distinguishing synthetic from real lesions, validating realism and anatomical plausibility. By delivering training-free, controllable synthesis of diverse lymph node shapes and textures, CAFusion enhances the utility of synthetic data for medical image analysis and downstream tasks.

Abstract

Lesion synthesis methods have made significant progress in generating large-scale synthetic datasets. However, existing approaches predominantly focus on texture synthesis and often fail to accurately model masks for anatomically complex lesions. Additionally, these methods typically lack precise control over the synthesis process. For example, perirectal lymph nodes, which range in diameter from 1 mm to 10 mm, exhibit irregular and intricate contours that are challenging for current techniques to replicate faithfully. To address these limitations, we introduce CAFusion, a novel approach for synthesizing perirectal lymph nodes. By leveraging Signed Distance Functions (SDF), CAFusion generates highly realistic 3D anatomical structures. Furthermore, it offers flexible control over both anatomical and textural features by decoupling the generation of morphological attributes (such as shape, size, and position) from textural characteristics, including signal intensity. Experimental results demonstrate that our synthetic data substantially improve segmentation performance, achieving a 6.45% increase in the Dice coefficient. In the visual Turing test, experienced radiologists found it challenging to distinguish between synthetic and real lesions, highlighting the high degree of realism and anatomical accuracy achieved by our approach. These findings validate the effectiveness of our method in generating high-quality synthetic lesions for advancing medical image processing applications.

Figures (3)

• Figure 1: Illustration of Controllable Perirectal Lymph Node Synthesis. Our method consists of three key steps. First, we train a network to generate the anatomical structure of perirectal lymph nodes, establishing a morphology prior. Second, we train an image inpainting network to construct a texture prior. Finally, we perform controllable synthesis by incorporating anatomical guidance during mask generation and textural guidance during texture generation.
• Figure 2: Anatomical Guidance (AG). This figure illustrates a framework for controlling 3D lymph node morphology, emphasizing both the global regulation of overall shape and the fine adjustment of surface details.
• Figure 3: Textural Guidance (TG). Leveraging the synthetic mask, we controllably generate perirectal lymph nodes on real background images. The figure demonstrates lymph nodes with varying Signal Intensity (SI) values, from low to high, where higher SI values enhance the contrast against the background.