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

Abstract

Purpose: This study aims to develop a machine learning-based model for the objective assessment of CT pulmonary angiography (CTPA) image quality.

Patients and methods: A retrospective analysis was conducted using data from 99 patients who underwent CTPA between March 2022 and January 2023, alongside two public datasets, FUMPE (21 cases) and CAD-PE (30 cases). In total, 150 cases from multiple centers were included in this analysis. The dataset was randomly split into a training set (105 cases) and a testing set (45 cases) in a 7:3 ratio. CT values and their standard deviations (SD) were measured in 11 specific regions of interest, and two radiologists independently assigned anonymous random scores to the images. The average of their subjective scores was used as the target output for the model, which was the mean opinion score (MOS) for image quality. Feature selection was performed using the Lasso algorithm and Pearson correlation coefficient, and a random forest regression model was constructed. Model performance was evaluated using mean square error (MSE), coefficient of determination (R²), Pearson linear correlation coefficient (PLCC), Spearman rank correlation coefficient (SRCC), and Kendall rank correlation coefficient (KRCC).

Results: After feature selection, three key features were retained: main pulmonary artery CT value, ascending aorta CT value, and the difference in noise values between the left and right main pulmonary arteries. The random forest regression model constructed achieved MSE, R2_score, PLCC, SRCC, and KRCC values of 0.2001, 0.6695, 0.8682, 0.8694, 0.7363, respectively, on the testing set.

Conclusion: This study successfully developed an interpretable machine learning-based model for the objective assessment of CTPA image quality. The model offers effective support for improving image quality control efficiency and precision. However, the limited sample size may affect the model's generalizability, so it's essential to conduct further research with larger datasets.