Author: Jiayun Chen, Shengqi Chen, Yuan Tang, Zilin Wang, Guohua Wu, Jianan Wu 👨🔬
Affiliation: Cancer Hospital Chinese Academy of Medical Sciences, Shenzhen Center, School of Electronic Engineering, Beijing University of Posts and Telecommunications, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College 🌍
Purpose:
To develop a novel no-reference image quality assessment (NRIQA) method for evaluating the effectiveness of image preprocessing in MRI-guided radiotherapy (MRIgRT), thereby enhancing clinical application precision and standardizing medical image quality assessment.
Methods:
A dataset of 106,000 MR images from 10 patients with liver metastasis, acquired using the Elekta Unity MR-LINAC system, was analyzed. The proposed NRIQA method integrates three key components: (1)Image Preprocessing: Enhancement of diagnostic feature visibility through optimized preprocessing techniques.(2)Feature Extraction and Directional Analysis: Utilization of Mean Subtracted Contrast Normalized (MSCN) coefficients across four directions to capture textural attributes and identify potential distortions.(3)Quality Index (QI) Calculation: Integration of features via Asymmetric Generalized Gaussian Distribution (AGGD) parameter estimation and K-means clustering to provide a comprehensive image quality measure.
Tumor tracking experiments were conducted using the Tracking-Learning-Detection (TLD) algorithm on both preprocessed and unprocessed images to validate the effectiveness of preprocessing. The QI was compared with conventional metrics (Contrast-to-Noise Ratio [CNR], Tenengrad gradient, and entropy) through statistical analysis to assess its performance and sensitivity.
Results:
Preprocessing significantly improved tumor tracking precision from 78.6% to 94.9% and recall from 7.4% to 76.5%, demonstrating the effectiveness of the proposed method. In distinguishing the difference before and after preprocessing, the QI outperformed conventional metrics, achieving improvements of 79.6 times over CNR, 6.5 times over Tenengrad gradient, and 1.7 times over entropy. These results highlight the superior sensitivity and performance of the QI in detecting image quality improvements.
Conclusion:
This study introduces a robust NRIQA method based on automated distortion recognition, offering a standardized quality control tool for MRIgRT. The proposed method enhances clinical application precision by improving tumor tracking accuracy and facilitating the standardization of medical image quality assessment, with significant implications for both clinical practice and research.