Deep learning-driven ultrasound equipment quality assessment with ATS-539 phantom data

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Ultrasound Technology

Background:

• Ultrasound imaging is crucial for disease detection, but image quality can be subjective and impact diagnosis.
• Poor ultrasound image quality increases the risk of misdiagnosis.
• Current quality assessments rely heavily on evaluator experience, leading to variability.

Purpose of the Study:

• To develop an objective, deep learning-based framework for assessing ultrasound image quality.
• To quantify image quality across critical parameters: 'Dead zone', 'Axial/lateral resolution', and 'Gray scale and dynamic range'.
• To establish a streamlined approach for consistent ultrasound equipment quality management.

Main Methods:

• A two-stage deep learning framework was developed using phantom data.
• Stage 1 automatically identified regions of interest for each quality parameter.
• Stage 2 utilized detection/classification models for quality evaluation, with results combined via logistic regression.

Main Results:

• The deep learning models achieved high AUC scores: 98.6% for 'Dead zone', 87.7% for 'Axial/lateral resolution', and 96.0% for 'Gray scale and dynamic range'.
• Analysis of guideline-compliant images showed excellent performance (AUCs: 98.2%, 92.8%, 100%).
• The framework demonstrated potential for quantitative and objective ultrasound image quality assessment.

Conclusions:

• Deep learning offers a powerful tool for objective ultrasound image quality evaluation.
• This framework enables consistent quality control and efficient, scoring-based assessment of ultrasound equipment.
• The study paves the way for improved diagnostic accuracy through standardized image quality management.