WING: A Window-Prior-Based Generative Network with Gated Inception for Cross-Modality CT Synthesis

2017-03-22 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical Image Synthesis · Depth: Expert, long

Summary

WING is a novel window-prior-based generative network designed for cross-modality CT synthesis, specifically addressing challenges in adaptive radiotherapy by generating CT volumes from MRI and CBCT. It reformulates CT intensity regression into multiple windowed representations (lung, soft tissue, bone) to handle high dynamic range and long-tailed distributions. The network integrates a Gated Inception Generator (GIG) for multi-window predictions, a Fuse-and-Refine Transformer (FRT) to aggregate and refine these outputs, and a joint adversarial training objective for enhanced realism. Evaluated on the large-scale SynthRAD2025 dataset, WING achieved state-of-the-art performance on both MRI-to-CT and CBCT-to-CT benchmarks. It significantly outperformed the previous MedNeXt model, demonstrating improved fidelity across heterogeneous attenuation regimes and supporting multi-anatomy synthesis with a single compact model.

Key takeaway

For AI Scientists and Machine Learning Engineers developing medical image synthesis models, adopting a window-prior-based approach for CT generation can significantly improve fidelity and geometric consistency. You should consider reformulating direct CT intensity regression into multiple windowed representations to better handle dynamic range and long-tailed distributions. This strategy, exemplified by WING, enhances detail across diverse anatomical structures, making your synthetic CTs more reliable for adaptive radiotherapy planning.

Key insights

CT synthesis benefits from window-prior-based regression to manage intensity distributions and improve structural fidelity.

Principles

• CT intensities are structure-deterministic and window-separable.
• Multi-kernel interactions enhance feature learning efficiency.
• Residual refinement improves coarse fused CT details.

Method

WING employs a Gated Inception Generator for multi-window predictions, a Fuse-and-Refine Transformer for differentiable soft fusion and residual refinement, and a joint adversarial objective for realism.

In practice

• Reformulate CT regression targets into windowed views.
• Use anisotropic kernels for thin, high-contrast structures.
• Apply differentiable soft fusion for smooth window blending.

Topics

• Synthetic CT
• Adaptive Radiotherapy
• Generative Adversarial Networks
• Medical Image Synthesis
• CT Windowing
• Transformers

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• WING: A Window-Prior-Based Generative Network with Gated Inception for Cross-Modality CT Synthesis — WING synthesizes CT from MRI/CBCT using windowed representations and a multi-component generative network for improved radiotherapy planning.
• Dual-Domain Equivariant Generative Adversarial Network for Multimodal CT-PET Synthesis — Dual-domain learning and rotational equivariance significantly enhance multimodal medical image synthesis accuracy and robustness.
• CONFLUX: A Latent Diffusion Model for 3D Chest-CT Synthesis with RL Post-Training — Latent diffusion with RL post-training significantly enhances controllable 3D medical image synthesis fidelity and faithfulness.
• Synthesize Realistic 3D Medical Images at Scale to Ship Pre‑Trained Models — NVIDIA's MAISI and NV-Generate frameworks provide scalable, open-source 3D medical image synthesis to overcome data bottlenecks.
• LUCID: Learned Undersampling-Adaptive Consistency-Guided Inference with Deterministic Flow Matching for Sparse-View CT Reconstruction — LUCID employs Flow Matching and sparsity-adaptive consistency guidance for robust sparse-view CT reconstruction, reducing artifacts and hallucinations.
• Parameter-Efficient Adaptation of SAM 3 for Automated ITV Generation from 4DCT Images — Parameter-efficient fine-tuning of SAM 3 with temporal coherence effectively automates ITV generation from 4DCT images using minimal data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.