ProSAC-CT: Progressive Spectral-Anatomical Co-Guided Multi-Stage Diffusion Model for Low-Dose CT Denoising

2026-07-02 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

ProSAC-CT, a progressive spectral-anatomical co-guided multi-stage diffusion model, addresses challenges in low-dose computed tomography (LDCT) denoising. LDCT reduces radiation exposure but introduces quantum noise, streak artifacts, and texture degradation, obscuring anatomical boundaries. Existing diffusion models often lack sufficient anatomical guidance and frequency-dependent recovery. ProSAC-CT integrates an anatomical-prior-guided conditioning (APGC) module for structural guidance, a residual frequency-domain decoupling stage (RFDDS) for frequency-aware representations, and a time-step-decoupling denoising decoder (TD3) to manage reverse-diffusion stages for anatomical stabilization, boundary refinement, and fine-detail recovery. Experiments on four LDCT degradation benchmarks demonstrated improved image fidelity, structural similarity, perceptual quality, and information preservation. Furthermore, downstream anatomical-region classification on Mayo-2020 indicated ProSAC-CT retains task-relevant anatomical information, supporting its practical utility.

Key takeaway

For medical imaging engineers or radiologists evaluating low-dose CT denoising solutions, ProSAC-CT offers a robust approach. Its multi-stage diffusion model, integrating anatomical and spectral guidance, significantly improves image fidelity and preserves critical boundary-sensitive anatomical details. You should consider ProSAC-CT for applications where maintaining diagnostic quality from LDCT scans is paramount, as it demonstrably retains task-relevant anatomical information for downstream analysis.

Key insights

ProSAC-CT uses co-guided multi-stage diffusion with anatomical and spectral priors to denoise low-dose CT images effectively.

Principles

• Low-dose CT denoising benefits from anatomical guidance.
• Frequency-aware representations improve image recovery.
• Decoupling denoising stages enhances detail preservation.

Method

ProSAC-CT employs APGC for structural guidance, RFDDS for frequency-aware representations, and TD3 to assign these to reverse-diffusion stages for stabilization, refinement, and detail recovery.

In practice

• Improve LDCT image quality for diagnosis.
• Preserve anatomical details in noisy scans.
• Enhance task-relevant information for classification.

Topics

• Low-Dose CT Denoising
• Diffusion Models
• Medical Imaging
• Anatomical Guidance
• Spectral Analysis
• Image Reconstruction

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

Related on AIssential

• CT-DegradBench: A Physics-Informed Benchmark for CT Degradation Detection and Severity Estimation — CT-DegradBench and SeSpeCT provide a unified benchmark and framework for detecting and estimating CT image degradation and severity.
• Prob-BBDM: a Probabilistic Brownian Bridge Diffusion Model for MRI sequence image-to-image translation — A probabilistic Brownian Bridge Diffusion Model efficiently synthesizes high-quality MRI sequences, preserving diagnostic information for clinical use.
• CA-GCL: Cross-Anatomy Global-Local Contrastive Learning for Robust 3D Medical Image Understanding — CA-GCL uses global-local contrastive learning and clinical-aware text augmentation to prevent representation collapse in 3D medical FVLP.
• [R] Seeing arxiv endorser (eess.IV or cs.CV) CT lung nodule AI validation preprint — CT slice thickness significantly impacts AI lung nodule detection sensitivity more than dose reduction.
• Multi-FRuGaL: Multimodal Flexible Redundancy-aware Decomposed Gated Learning for Cancer Diagnosis and Prognosis — Multi-FRuGaL robustly diagnoses cancer from incomplete multimodal medical data by separating redundant from complementary signals.
• Margin-Consistent Deep Subtyping of Invasive Lung Adenocarcinoma via Perturbation Fidelity in Whole-Slide Image Analysis — Margin consistency with Perturbation Fidelity enhances deep learning robustness and accuracy for lung adenocarcinoma subtyping.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.