Problem
Research questions and friction points this paper is trying to address.
Ultrasound Image Restoration
Image Enhancement
Unaligned Data
Compound Degradation
Style-Driven Data Synthesis and Degradation-Aware Enhancement for Ultrasound Image Restoration
🤖 AI Summary
Handheld ultrasound images face the dual challenge of severe compound degradation and a lack of pixel-aligned training data. This work proposes a two-stage enhancement framework to address these issues. First, an unsupervised style transfer approach based on cycle-consistent generative adversarial networks is employed to construct a high-quality aligned dataset. Subsequently, building upon the PiSA-SR architecture, a dual-degradation-guided low-rank adaptation (LoRA) correction mechanism is designed for image enhancement. Evaluated on the USenhance2023 dataset, the proposed method achieves a 16.7% improvement in the Fréchet Inception Distance (FID) metric, with output distributions highly consistent with those of authentic high-quality images. These results demonstrate that this framework provides an effective solution for ultrasound image restoration in resource-constrained scenarios.
📝 Abstract
Low-cost handheld ultrasound devices can be widely deployed compared to professional hospital ultrasound machines. However, their images suffer from compound degradation that can mislead clinical judgment. Motivated by this observation, mapping handheld low-quality (LQ) to hospital high-quality (HQ) images has been considered a valuable research question. Conventionally, the mapping requires pixel-aligned LQ-HQ pairs. This requirement is unsatisfactory in practical scenarios because real scans at different times are never pixel-aligned. This paper addresses the challenge with a two-stage framework. The first stage generates pixel-aligned LQ-HQ datasets, and the second stage trains an enhancement model that improves LQ images. The first stage trains a cycle-consistent style-transfer model on unaligned real LQ-HQ pairs to learn a HQ-to-LQ model. Then, the model transforms real HQ images into pixel-aligned LQ images. Based on the dataset generated by the first stage, the second stage uses the Dual Degradation-Guided (DDG) Low-Rank Adaptation (LoRA) method to fine-tune an LQ-to-HQ model based on aligned pairs. In this stage, the model is based on the well known PiSA-SR framework but inserts a degradation-conditioned correction matrix. Experimental results on the USenhance2023 dataset show that the FID metric is improved by 16.7% over the strongest baseline while other metrics indicate that our enhanced outputs are well aligned with the real HQ distribution. The source code of our method is available at https://github.com/Jason0411202/DDG_LoRA.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Style Transfer
Data Synthesis
Dual Degradation-Guided LoRA
Ultrasound Image Restoration
Cycle Consistency
🔎 Similar Papers
No similar papers found.
