Enhanced Low-Dose CT Image Reconstruction by Domain and Task Shifting Gaussian Denoisers

2024-03-06

AI Summary

To address the challenges of high noise levels and the trade-off between reconstruction quality and computational efficiency in low-dose CT (LDCT) imaging, this paper proposes an efficient two-stage reconstruction framework. First, filtered back-projection (FBP) generates an initial reconstruction; second, a general-purpose Gaussian denoiser—pretrained on natural images and subsequently fine-tuned via domain adaptation (natural grayscale → LDCT) and task adaptation (Gaussian denoising → LDCT enhancement)—enhances the FBP output. Notably, the method requires no additional pretraining, is architecture-agnostic, and significantly reduces computational overhead. When fine-tuned on the LoDoPaB-CT dataset, our approach achieves top-ranked average performance in the associated challenge. Under limited-data settings, it outperforms conventional two-stage methods and approaches the quality of unrolled iterative networks, while offering substantially faster inference than learned primal-dual and similar deep iterative schemes.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Security and Privacy: Privacy-enhancing technologies

Abstract

Computed tomography from a low radiation dose (LDCT) is challenging due to high noise in the projection data. Popular approaches for LDCT image reconstruction are two-stage methods, typically consisting of the filtered backprojection (FBP) algorithm followed by a neural network for LDCT image enhancement. Two-stage methods are attractive for their simplicity and potential for computational efficiency, typically requiring only a single FBP and a neural network forward pass for inference. However, the best reconstruction quality is currently achieved by unrolled iterative methods (Learned Primal-Dual and ItNet), which are more complex and thus have a higher computational cost for training and inference. We propose a method combining the simplicity and efficiency of two-stage methods with state-of-the-art reconstruction quality. Our strategy utilizes a neural network pretrained for Gaussian noise removal from natural grayscale images, fine-tuned for LDCT image enhancement. We call this method FBP-DTSGD (Domain and Task Shifted Gaussian Denoisers) as the fine-tuning is a task shift from Gaussian denoising to enhancing LDCT images and a domain shift from natural grayscale to LDCT images. An ablation study with three different pretrained Gaussian denoisers indicates that the performance of FBP-DTSGD does not depend on a specific denoising architecture, suggesting future advancements in Gaussian denoising could benefit the method. The study also shows that pretraining on natural images enhances LDCT reconstruction quality, especially with limited training data. Notably, pretraining involves no additional cost, as existing pretrained models are used. The proposed method currently holds the top mean position in the LoDoPaB-CT challenge.

Problem

• Improves low-dose CT image reconstruction quality.
• Combines simplicity of two-stage methods with advanced quality.
• Uses pretrained Gaussian denoisers for enhanced LDCT images.

Innovation

• Combines two-stage simplicity with advanced quality
• Uses pretrained Gaussian denoisers for LDCT enhancement
• Fine-tunes natural image models for medical imaging