CT-DegradBench: A Physics-Informed Benchmark for CT Degradation Detection and Severity Estimation

2026-05-19 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning · Depth: Expert, long

Summary

CT-DegradBench is a new dataset and benchmark designed to systematically evaluate CT image degradation detection and severity estimation. It addresses the limitations of existing datasets, which focus on isolated restoration tasks and rely on image quality metrics with limited clinical validity. CT-DegradBench includes paired reference and degraded CT images generated using physics-informed models, covering five common degradation types (noise, blur, streaking, aliasing, metal artifacts) across four calibrated severity levels (L0-L3), as well as radiologist-informed mixed-artifact settings. Alongside this, the paper introduces SeSpeCT (Semantic-Spectral CT degradation estimation), a training-free framework that combines semantic priors from medical vision-language models with frequency-domain cues to jointly predict artifact type and severity. SeSpeCT constructs a semantic quality axis in the multimodal embedding space using radiology-informed text prompts and integrates spectral features to capture degradation-specific frequency patterns, outperforming baselines in both single and mixed degradation scenarios.

Key takeaway

For Machine Learning Engineers developing CT image enhancement or quality assessment models, CT-DegradBench offers a standardized, comprehensive benchmark to evaluate model performance across diverse, clinically relevant degradation types and severity levels. You should consider integrating semantic and spectral features, as demonstrated by SeSpeCT, to improve joint artifact detection and severity estimation, moving beyond traditional IQA metrics for more robust clinical applicability.

Key insights

CT-DegradBench and SeSpeCT provide a unified benchmark and framework for detecting and estimating CT image degradation and severity.

Principles

• Physics-informed degradation models enable controlled artifact simulation.
• Multimodal embeddings can define semantic quality axes.
• Frequency-domain features complement semantic cues for artifact analysis.

Method

SeSpeCT combines a semantic quality branch (using BioMedCLIP and prompt-derived quality axis) with a spectral branch (Fourier transform, convolutional encoder, high-frequency energy ratio) for fused multi-task prediction.

In practice

• Use physics-informed models for realistic CT artifact generation.
• Employ medical VLMs with text prompts for quality assessment.
• Integrate spectral analysis for degradation-specific frequency patterns.

Topics

• CT Degradation Detection
• Severity Estimation
• CT-DegradBench Dataset
• SeSpeCT Framework
• Vision-Language Models

Code references

• yousranb/CT-DEGRADBENCH

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

Related on AIssential

• Benchmarks and methods for 3D medical image retrieval — The 3D-MIR benchmark and methods advance medical image retrieval by providing standardized evaluation and multi-modal search strategies.
• Generative Deep Learning for Computational Destaining and Restaining of Unregistered Digital Pathology Images — Preprocessing-based domain adaptation enables cGAN models to generalize for computational pathology across institutions without retraining.
• Cross-Source Supervision for Bone Infection Segmentation in Dual-Modality PET-CT — Decoupled dual-source learning with PET-CT fusion improves bone infection segmentation by accommodating diverse clinical annotation philosophies.
• PEFT-MedSAM: Efficient Fine-Tuning of Medical Foundation Models for Explainable Skin Lesion Segmentation — PEFT-MedSAM efficiently fine-tunes medical foundation models for skin lesion segmentation by training only the mask decoder.
• 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.
• Multi-Contrast MRI Motion Correction via Parameter-Informed Disentanglement and Adaptive Experts — A unified MRI motion correction framework disentangles contrast and adapts to severity for robust, generalizable artifact removal.

Open in AIssential →

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