A study of the value of three-dimensional T 1WI using different acceleration methods in the application of brain region segmentation
10.3760/cma.j.cn112149-20240310-00115
- VernacularTitle:不同加速方式的三维T 1WI对脑区分割应用价值研究
- Author:
Gang YAO
1
;
Zhuqing ZHOU
;
Feng SHI
;
Zehong CAO
;
Xiaopeng SONG
;
Weijun ZHANG
;
Wenwen SHEN
Author Information
1. 宁波大学附属康宁医院（宁波市康宁医院）精神科，宁波 315201
- Keywords:
Magnetic resonance imaging;
Parallel imaging;
Compressed sensing technique;
Artificial intelligence;
Segmentation of brain areas
- From:
Chinese Journal of Radiology
2024;58(10):1006-1014
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of three-dimensional (3D) T 1WI structural images using different acceleration methods including parallel acquisition technique, joint compressed sensing (uCS) technique, and artificial intelligence-assisted compressed sensing (ACS) technique for brain region segmentation. Methods:In this cross-sectional study, fifty patients (female: n=25, age range: 13 to 87 years old) at Corning Hospital of Ningbo University from July to September 2023 were prospectively and consecutively collected. All the subjects underwent brain MRI. Six groups of 3D T 1WI structural images were obtained using different acceleration technique and parameters, including 3D T 1WI without acceleration factor (3D-T 1WI group), 3D T 1WI with parallel acquisition technique with acceleration factor 3 (3D-T 1WI-PI-3 group), 3D T 1WI with uCS technique with acceleration factor 4.5 and 6.9 (3D-T 1WI-uCS-4.5 group, 3D-T 1WI-uCS-6.9 group), 3D T 1WI by ACS technique with acceleration factors of 3 and 5 (3D-T 1WI-ACS-3 group, 3D-T 1WI-ACS-5 group). T 2WI fluid-attenuated inversion recovery (FLAIR) images were also acquired. Subjective scores (cerebral grey matter and white matter clarity scores, clarity scores of cerebral white matter degeneration lesions in relation to the surrounding white matter, and Gibbs artifact scores) and objective metrics [signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), cerebrospinal fluid signal homogeneity, peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and natural image quality evaluator (NIQE)] were used to evaluate image quality in different groups. Totally 109 brain regions were segmented and volumes were measured using the uAI Research Portal image analysis tool. Kappa or intraclass correlation coefficient ( ICC) was used to evaluate the agreement of subjective and objective evaluation indexes between the 3D-T 1WI-PI-3 group, 3D-T 1WI-uCS-4.5 group, 3D-T 1WI-uCS-6.9 group, 3D-T 1WI-ACS-3 group, 3D-T 1WI-ACS-5 group, and 3D-T 1WI group. Kappa or ICC value>0.70 was considered as good agreement. Results:The acquisition time for the 3D-T 1WI group, 3D-T 1WI-PI-3 group, 3D-T 1WI-uCS-4.5 group, 3D-T 1WI-uCS-6.9 group, 3D-T 1WI-ACS-3 group, and 3D-T 1WI-ACS-5 group were 527, 204, 169, 95, 133, 90 s, respectively. Subjective evaluation showed that the 3D-T 1WI-uCS-4.5, 3D-T 1WI-ACS-3, and 3D-T 1WI-ACS-5 groups had excellent agreement with the 3D-T 1WI group in terms of the distribution of cases of cerebral grey matter and white matter clarity scores, respectively (all Kappa value=1.000); The distribution of cases of clarity score of cerebral white matter lesions and surrounding white matter in the 3D-T 1WI-PI-3 group, 3D-T 1WI-uCS-4.5 group, and 3D-T 1WI-ACS-3 group were in good agreement with that of the 3D-T 1WI group ( Kappa values of 0.775, 0.701, and 0.777, respectively); the distribution of the number of cases of the Gibbs artifact score of the 3D-T 1WI-uCS-4.5, 3D-T 1WI-ACS-3, and 3D-T 1WI-ACS-5 groups was in good agreement with the 3D-T 1WI group (all Kappa value=1.000). Objective evaluation showed the CNR of the images in the 3D-T 1WI-PI-3, 3D-T 1WI-uCS-4.5, and 3D-T 1WI-uCS-6.9 groups were in good agreement with those of the 3D-T 1WI group ( ICC of 0.720, 0.759, and 0.752, respectively); PSNR and SSIM were in good agreement among the 3D-T 1WI-PI-3 group, 3D-T 1WI-uCS-4.5 group, 3D-T 1WI-uCS-6.9 group, 3D-T 1WI-ACS-3 group, and 3D-T 1WI-ACS-5 group (PSNR: ICC=0.854; SSIM: ICC=0.851). NIQE of 3D-T 1WI-PI-3 group, 3D-T 1WI-uCS-4.5 group, and 3D-T 1WI-ACS-3 group images were in good agreement with the 3D-T 1WI group ( ICC value of 0.866, 0.727, 0.753, respectively). The ICC values of the volume of each segmented brain region among the 3D-T 1WI-PI-3, 3D-T 1WI-uCS-4.5, 3D-T 1WI-uCS-6.9, 3D-T 1WI-ACS-3, 3D-T 1WI-ACS-5 group and the 3D-T 1WI group images showed decreased in order (all ICC≥0.62). Conclusions:The uCS and ACS techniques used in 3D-T 1WI show high agreement with 3D-T 1WI in terms of brain segmentation. The application of these accelerating techniques can significantly shorten the acquisition time with obtaining images with good image quality, displaying great value.
