基于深度学习生成的磁共振图像的临床验证研究  

Clinical Validation Study of Deep Learning-Generated Magnetic Resonance Images

作　　者：伏广东 彭立凤 张志浩 项磊 王泷 何健[1] FU Guangdong;PENG Lifeng;ZHANG Zhihao;XIANG Lei;WANG Long;HE Jian(The Gulou School of Clinical Medicine,Nanjing Medical University,Nanjing,210000;Subtle Medical Inc.,Shanghai,200062)

机构地区：[1]南京医科大学鼓楼临床医学院,南京市210000 [2]深智透医疗科技发展(上海)有限责任公司,上海市200062

出　　处：《中国医疗器械杂志》2024年第5期493-497,共5页Chinese Journal of Medical Instrumentation

摘　　要：该研究采用基于深度学习的图像生成算法,从磁共振图像的矢状位T1WI和T2WI序列生成伪矢状位STIR序列。以真实采集的STIR序列为金标准,从主观和客观两方面评价生成序列的图像质量及诊断效果。主观方面,由两位医师对图像进行评分。客观方面,选取5种组织的感兴趣区域,计算信噪比和对比度噪声比来衡量图像质量。此外,采用平均绝对误差、峰值信噪比、结构相似度和相关系数来分析生成的STIR与金标准图像的相关性;使用Bland-Altman图分析像素值的一致性。研究结果表明,深度学习生成的STIR序列在图像质量和临床诊断上与金标准一致甚至超越金标准,且能够缩短扫描时间,提高影像扫描效率,因此具有应用于临床实践的前景。This research utilizes a deep learning-based image generation algorithm to generate pseudo-sagittal STIR sequences from sagittal T1WI and T2WI MR images.The evaluations include both subjective assessments by two physicians and objective analyses,measuring image quality through SNR and CNR in ROIs of five different tissues.Further analyses,including MAE,PSNR,SSIM,and COR,establish a strong correlation between the generated STIR sequences and the gold standard,with Bland-Altman analysis indicating pixel consistency.The findings indicate that the deep learning-generated STIR sequences not only align with but potentially surpass the gold standard in terms of image quality and clinical diagnostic capabilities.Moreover,the approach demonstrates promise for clinical implementation,offering reduced scan time and enhanced imaging efficiency.

关 键 词：磁共振图像 深度学习 图像生成 脊柱 

分 类 号：R445.2[医药卫生—影像医学与核医学] TH77[医药卫生—诊断学]