Magnetic resonance shoulder imaging using deep learning-based algorithm.
Eur Radiol
; 33(7): 4864-4874, 2023 Jul.
Article
em En
| MEDLINE
| ID: mdl-36826500
ABSTRACT
OBJECTIVE:
To investigate the feasibility of deep learning-based MRI (DL-MRI) in its application in shoulder imaging and compare its performance with conventional MR imaging (non-DL-MRI).METHODS:
This retrospective study was approved by the local ethics committee. Seventy consecutive patients who had been examined with both DL-MRI and non-DL-MRI were enrolled for the image quality and lesion diagnosis comparison. Another 400 patients had been examined only with DL-MRI. Their images' quality was assessed by 20 radiologists using a satisfaction survey. The Kendall W test was performed to assess interobserver agreement. The Wilcoxon test was performed to compare the image quality. For lesion diagnosis, the interobserver and interstudy agreement were evaluated by kappa analysis.RESULTS:
The scan time of DL-MRI (6 min 1 s) was nearly 50% decreased compared with that of non-DL-MRI (11 min 25 s). The image quality was higher in both PDWI (4.85 ± 0.31 for DL, and 4.73 ± 0.29 for non-DL) and T2WI (4.95 ± 0.2 for DL, and 4.74 ± 0.41 for non-DL) of DL-MRI. Good interobserver agreement was found for the image quality of all the MR sequences on both DL-MRI (Kendall W 0.588~0.902) and non-DL-MRI (Kendall W 0751~0.865). Both the SNRs and |CNR| were significantly higher in PDWI and T2WI of DL-MRI. High interobserver and interstudy agreements for the lesions in non-DL-MRI and DL-MRI (kappa value = 0.913 to 1.000) were observed. The results of the image quality satisfaction survey in 400 patients receiving DL-MRI in the shoulder obtained 5 scores among all the radiologists.CONCLUSION:
Shoulder DL-MRI can greatly reduce the scan time, while improve imaging quality of PDWI and T2WI compared to non-DL-MRI. KEY POINTS ⢠Shoulder 2D DL-MRI can greatly reduce the whole scan time and improve imaging quality of both PDWI and T2WI compared to conventional parallel MRI. ⢠Shoulder 2D DL-MRI could be a clinical routine with greatly improved work efficiency in the future.Palavras-chave
Texto completo:
1
Coleções:
01-internacional
Base de dados:
MEDLINE
Assunto principal:
Ombro
/
Aprendizado Profundo
Tipo de estudo:
Observational_studies
/
Prognostic_studies
Limite:
Humans
Idioma:
En
Revista:
Eur Radiol
Assunto da revista:
RADIOLOGIA
Ano de publicação:
2023
Tipo de documento:
Article
País de afiliação:
China