Machine Learning-Based Objective Evaluation Model of CTPA Image Quality: A Multi-Center Study.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Computed tomography pulmonary angiography (CTPA) is crucial for diagnosing pulmonary embolism.
• Objective assessment of CTPA image quality is essential for accurate diagnosis and treatment.
• Current methods for image quality assessment are often subjective and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning-based model for the objective assessment of CTPA image quality.
• To identify key image features that correlate with subjective image quality scores.
• To improve the efficiency and precision of CTPA image quality control.

Main Methods:

• A retrospective analysis of 150 CTPA cases from multiple centers was performed.
• A random forest regression model was trained using CT values and radiologist scores (Mean Opinion Score - MOS).
• Feature selection was conducted using Lasso algorithm and Pearson correlation, with performance evaluated using MSE, R², PLCC, SRCC, and KRCC.

Main Results:

• The developed model identified three key features: main pulmonary artery CT value, ascending aorta CT value, and inter-arterial noise difference.
• The random forest model achieved strong performance with MSE of 0.2001, R² of 0.6695, PLCC of 0.8682, SRCC of 0.8694, and KRCC of 0.7363 on the testing set.
• The model demonstrated interpretable results, correlating objective image features with subjective quality assessments.

Conclusions:

• An interpretable machine learning model for objective CTPA image quality assessment was successfully developed.
• The model provides effective support for enhancing image quality control efficiency and precision.
• Further research with larger datasets is recommended to improve the model's generalizability.