Evaluation of image quality of deep learning-based reconstruction of prostate T 2WI and diagnostic performance for prostate cancer in transition zone
10.3760/cma.j.cn112149-20230703-00452
- VernacularTitle:基于深度学习重建的前列腺T 2WI的图像质量和对移行带前列腺癌的诊断效能
- Author:
Bowen YANG
1
;
Hao CHENG
;
Ming LIU
;
Huimin HOU
;
Miao WANG
;
Chen ZHANG
;
Chunmei LI
;
Min CHEN
Author Information
1. 北京医院放射科 国家老年医学中心 中国医学科学院老年医学研究院，北京 100730
- Keywords:
Prostate;
Magnetic resonance imaging;
Image quality;
Deep learning
- From:
Chinese Journal of Radiology
2023;57(11):1208-1214
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To evaluate the image quality of prostate T 2WI reconstructed based on deep learning (deep learning T 2WI) and the diagnostic performance for prostate cancer (PCa) in the transition zone. Methods:Totally 79 patients who underwent prostate MRI for elevated prostate specific antigen from December 2020 to September 2022 were prospectively consecutively collected from Beijing Hospital. Scan sequences included axial standard T 2WI, deep learning T 2WI, and diffusion-weighted imaging. The scan time was recorded. The image quality was scored subjectively including image quality, diagnostic confidence, noise level, artifacts, clarity and lesion detectability. For objective evaluation of image quality, signal-to-noise ratio (SNR) and contrast signal-to-noise ratio (CNR) were calculated. Two-parameter MRI prostate imaging reporting and data system version 2.1 (PI-RADS v2.1) scoring was performed for transition zone lesions using deep learning T 2WI and standard T 2WI, respectively. The subjective and objective image quality evaluation metrics for deep learning T 2WI and standard T 2WI were compared using the Wilcoxon signed-rank test. For transition zone lesions, the diagnostic performance of PI-RADS scores with deep learning T 2WI and standard T 2WI for PCa was evaluated by the receiver operating characteristic curve based on the lesion (all lesions in the transition zone) and the patient (the most malignant lesions in the transition zone), respectively, using the pathologic results as the gold standard. The area under the curve (AUC) was compared using the DeLong test. Results:Deep learning T 2WI significantly reduced the examination time by 64.6.%, from 4 min 37 s to 1 min 38 s. The scores of subjective image quality of deep learning T 2WI and standard T 2WI all were 5 (4, 5). The differences in image quality and lesion detectability were statistically significant ( Z=-2.32, -2.36, P=0.020, 0.018), and the differences of all other image quality evaluation metrics were not statistically significant ( P>0.05). The SNR of deep learning T 2WI and standard T 2WI were 17.11 (14.09, 21.92) and 9.15 (7.16, 11.17), with a statistically significant difference ( Z=-7.72, P<0.001). The CNR of deep learning T 2WI and standard T 2WI were 20.78 (13.42, 31.42) and 11.05 (7.82, 16.25), with a statistically significant difference ( Z=-7.54, P<0.001). Based on the lesion (40 PCa and 48 benign lesions), the AUC of the two-parameter PI-RADS score with deep learning T 2WI and standard T 2WI for diagnosing PCa in the transition zone were 0.915 (95%CI 0.856-0.975) and 0.916 (95%CI 0.857-0.976), without statistically significant difference ( Z=0.03, P=0.973). Based on the patient (33 PCa and 46 benign patients), the AUC of the two-parameter PI-RADS score with deep learning T 2WI and standard T 2WI were 0.921 (95%CI 0.857-0.984) and 0.939 (95%CI 0.886-0.992), without statistically significant difference ( Z=0.59, P=0.558). Conclusions:Compared with standard T 2WI, deep learning T 2WI of the prostate reduces scanning time while maintaining image quality and has comparable diagnostic performance for PCa in the transition zone.
