Automated Quality Assessment of Medical Images in Echocardiography Using Neural Networks with Adaptive Ranking and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate medical image quality is vital for disease diagnosis and treatment.
• Current automated quality assessment methods using neural networks primarily focus on pixel distortion, neglecting semantic information.
• This limitation impacts the reliability of automated medical image analysis.

Purpose of the Study:

• To develop a novel neural network model for automated medical image quality assessment.
• To address both pixel and semantic distortions for more comprehensive quality evaluation.
• To improve the accuracy and clinical utility of automated image quality assessment.

Main Methods:

• Introduced a novel neural network model for automated image quality assessment.
• Incorporated an adaptive ranking mechanism with contrast sensitivity weighting for pixel distortion.
• Integrated a structure-aware learning module using graph neural networks to analyze semantic structures.

Main Results:

• The proposed model demonstrated superior performance compared to existing leading models on two ultrasound imaging datasets.
• The structure-aware module effectively captured relationships between semantic content and image quality.
• The model achieved enhanced detection of minor variances in similar images.

Conclusions:

• The novel neural network model offers a more robust approach to automated medical image quality assessment.
• The integration of pixel and semantic analysis improves diagnostic and treatment accuracy.
• The model's seamless integration supports real-time clinical decision-making.