DAM: Degradation-Aware Model for Ultrasound Image Quality Assessment.

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Ultrasound image quality assessment (IQA) faces challenges due to intertwined semantic content and quality information (e.g., blurring, shadows).
• Subtle quality variations in ultrasound images complicate accurate IQA and can lead to biased results.
• Existing methods often fail to adequately address fine-grained quality inconsistencies.

Purpose of the Study:

• To develop a novel degradation-aware model (DAM) for accurate ultrasound IQA.
• To disentangle quality-related representations from semantic content in ultrasound images.
• To enhance the model's ability to perceive and learn subtle quality variations.

Main Methods:

• Proposed a degradation-aware model (DAM) incorporating degradation-derived augmentation (DDA) to synthesize appearance changes based on clinical concerns.
• Introduced fine-grained degradation learning (FGDL) to distinguish between images with diminishing quality inconsistencies.
• Developed a universal boundary acquisition operator (UBAO) to standardize ultrasound images and suppress redundant information.

Main Results:

• The proposed DAM model achieved superior performance compared to 14 baseline methods on an in-house ultrasound dataset.
• Achieved a peak signal-to-noise ratio (PLCC) of 0.760 and a Spearman rank-order correlation coefficient (SROCC) of 0.766.
• Demonstrated effective disentanglement of quality and semantic information and improved awareness of quality nuances.

Conclusions:

• The novel DAM effectively addresses the challenges in ultrasound IQA by disentangling semantic content and quality information.
• The proposed methods (DDA, FGDL, UBAO) significantly enhance the accuracy and robustness of ultrasound image quality assessment.
• The developed model offers a promising solution for reliable and standardized ultrasound image quality evaluation in clinical practice.