中国中西医结合影像学杂志2025，Vol.23Issue(5)：634-639,6.DOI:10.3969/j.issn.1672-0512.2025.05.020
深度学习在踝关节软骨MRI BLADE序列成像中的应用
Application of deep learning in ankle cartilage MRI by using BLADE sequence
摘要
Abstract
Objective:To compare the application of artificial intelligence deep learning by the super resolution enhancement(SRE)algorithm and the denoising enhancement(DNE)algorithm on ankle cartilage MRI by using BLADE sequence.Methods:15 patients with ankle cartilage injury(patient group)and 15 volunteers(control group)were enrolled.Two groups were first approved for water film imaging by MRI,and then two different K-space BLADE sequences imaging were performed(with coverage rates of 50.5%and 100%,respectively).All MRI images were deeply learned by SRE algorithm and DNE algorithms.SNR values between two kinds of BLADE sequences water film imaging and SRE,DNE images were compared.The ankle cartilage SNR and CNR(cartilage/subchondral bone,cartilage/synovial fluid)between two kinds of BLADE sequences and SRE,DNE were compared in the control group were compared.The imaging quality subjective scores conducted by 2 radiologists between two kinds of BLADE sequences and SRE,DNE after deep learning in patient group were compared,including the details of non-injured cartilage,diagnostic value for cartilage injury,noise and artifacts.Results:The scan time of BLADE-50.5%sequence was only 51.25%of that of BLADE-100%sequence.In water film images,SNR values of BLADE-50.5%and BLADE-100%images were 38.37(23.02,53.06),54.91(45.18,60.47).After deep learning,SNR values of BLADE-50.5%-SRE and BLADE-50.5%-DNE images were 54.60(25.10,74.54),61.26(26.75,84.94),respectively,with no statistical differences compared with BLADE-100%water film images(both P＞0.05).Among the six images in the control group,the difference in the cartilage SNR was statistically significant(P<0.001),and pairwise comparisons showed that the differences between the two original images and the BLADE-50.5%-SRE,BLADE-100%-SRE,BLADE-100%-DNE images were all statistically significant(all P<0.05).The difference in the cartilage/subchondral bone CNR in the six images of the control group was statistically significant(P<0.001),and pairwise comparisons showed that the differences between the two original images and the BLADE-50.5%-SRE,BLADE-100%-SRE,BLADE-100%-DNE images were all statistically significant(all P<0.05).The differences in the cartilage/synovial fluid CNR in the six images of the control group was statistically significant(P<0.001),and pairwise comparisons showed that the differences between the two original images and the four images after deep learning were statistically significant(all P<0.05).In the patient group,the subjective scores of SRE and DNE images after deep learning were all higher than their original images in three aspects,the three-layer structure of non-injured cartilage region,the diagnostic characteristics of injured cartilage region,the noise and motion artifacts(all P<0.05),and the BLADE-100%-SRE images had the highest scores,followed by BLADE-50.5%-SRE images.Conclusions:Deep learning can help to obtain ankle cartilage images with better quality,SRE has higher SNR and CNR and the best performance in the subjective judgment of ankle cartilage injury.关键词
磁共振成像/深度学习/踝关节/BLADE序列/软骨损伤Key words
Magnetic resonance imaging/Deep learning/Ankle/BLADE sequences/Osteochondral lesion引用本文复制引用
孙岩,邹月芬,庄启湘,刘可夫..深度学习在踝关节软骨MRI BLADE序列成像中的应用[J].中国中西医结合影像学杂志,2025,23(5):634-639,6.基金项目
苏州市卫生健康委员会-苏州市临床重点病种诊疗技术专项项目(LCZX202212). （LCZX202212）
