HiSin: Efficient High-Resolution Sinogram Inpainting via Resolution-Guided Progressive Inference

2025-06-10

AI Summary

High-resolution sinogram inpainting is critical for CT reconstruction, yet conventional diffusion models face prohibitive memory and computational overhead, hindering direct application. To address this, we propose a resolution-guided progressive diffusion framework that jointly optimizes global structural modeling and local detail restoration. Our method introduces two key innovations: (i) a frequency-aware patch skipping mechanism that prioritizes high-frequency regions requiring refinement, and (ii) a structure-adaptive step allocation strategy that dynamically assigns sampling steps per patch based on local structural complexity. It further integrates multi-scale inference, frequency-sensitive patch scheduling, dynamic step-length optimization, and localized high-resolution patch processing. Experiments demonstrate a 31.25% reduction in peak GPU memory usage and an 18.15% decrease in inference latency, while maintaining state-of-the-art inpainting accuracy across multiple datasets, resolutions, and mask configurations—effectively overcoming the efficiency bottleneck in high-resolution sinogram restoration.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Content-based information diffusion

Abstract

High-resolution sinogram inpainting is essential for computed tomography reconstruction, as missing high-frequency projections can lead to visible artifacts and diagnostic errors. Diffusion models are well-suited for this task due to their robustness and detail-preserving capabilities, but their application to high-resolution inputs is limited by excessive memory and computational demands. To address this limitation, we propose HiSin, a novel diffusion based framework for efficient sinogram inpainting via resolution-guided progressive inference. It progressively extracts global structure at low resolution and defers high-resolution inference to small patches, enabling memory-efficient inpainting. It further incorporates frequency-aware patch skipping and structure-adaptive step allocation to reduce redundant computation. Experimental results show that HiSin reduces peak memory usage by up to 31.25% and inference time by up to 18.15%, and maintains inpainting accuracy across datasets, resolutions, and mask conditions.

Problem

• Efficient high-resolution sinogram inpainting for CT reconstruction
• Reducing memory and computational demands of diffusion models
• Maintaining accuracy while improving efficiency in inpainting tasks

Innovation

• Resolution-guided progressive inference for efficiency
• Frequency-aware patch skipping reduces computation
• Structure-adaptive step allocation enhances performance