Deep Learning Technology for Rapid Morphological and Quantitative Imaging of Knee Pathology

用于膝关节病理学快速形态学和定量成像的深度学习技术

• 批准号: 10630920
• 负责人: Fang Liu
• 金额: $ 35.76万
• 依托单位: MASSACHUSETTS GENERAL HOSPITAL
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-06-01 至 2027-05-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10630920
• 关键词: 3-Dimensional; Acceleration; Adult; Advanced Development; Body part; Brain; Cartilage; Clinical; Development; Early Diagnosis; Economic Burden; Evaluation; Foundations; General Population; Goals; Health Care Costs; Health Personnel; Healthcare; High Prevalence; Image; Imaging Techniques; Incidence; Investigation; Joints; Knee; Knee Injuries; Knee Osteoarthritis; Knee joint; Magnetic Resonance Imaging; Maps; Medicare; Methods; Modeling; Morphology; Musculoskeletal; Noise; Pathology; Patients; Peripheral; Population; Prevalence; Protocols documentation; Public Health; Relaxation; Research; Resolution; Sampling; Scanning; Schedule; Signal Transduction; Slice; Societies; Structure; Techniques; Technology; Time; Tissues; Training; United States; Validation; Vertebral column; care burden; clinical imaging; clinical practice; deep learning; imaging modality; improved; joint destruction; knee pain; learning strategy; neural network; novel; operation; prevent; quantitative imaging; rapid technique; reconstruction; two-dimensional

PROJECT SUMMARY The high prevalence of knee pain in the general population has presented an immense challenge to public health, with significant health care and economic burden to our society. Magnetic resonance imaging (MRI) is the imaging modality of choice to evaluate patients with knee pain. Indeed, peripheral joints rank third as the most frequent body parts imaged using MRI, with the knee being by far the most common joint evaluated. Given the rise of the number of knee MRI examinations over the next decade with the increasing incidence of knee injuries and the increasing prevalence of knee osteoarthritis, there is an urgent clinical need to reduce the economic burden of knee MRI, with the most direct approach being to decrease the overall time required to perform the MRI examination. Over the past decade, multiple techniques have been attempted to accelerate knee MRI including parallel imaging, compressed sensing, multi-slice acquisition, and three-dimensional isotropic resolution imaging. However, all current methods have limitations, including decreased signal-to-noise ratio, image blurring, incompatibility to present necessary tissue contrasts, and inability to evaluate all joint structures. Lack of appropriate acceleration methods also prevents quantitative MRI such as T2 relaxation time mapping from being used clinically, despite its evident value for detecting early signs of joint degeneration. This application aims to develop novel rapid acquisition and reconstruction techniques to maximize MR scanner efficiency, improve imaging management, and automate scanning workflow, with the final goal of reducing the economic burden of knee MRI and facilitating clinical imaging operation. Our proposed new methods will be based on developing advanced deep learning reconstruction, combined with novel rapid image acquisition and automatic processing, all of which are pioneered by our research group. We propose developing, optimizing, and evaluating a rapid 5-minute knee MRI protocol consisting of all clinical sequences and additional T2 mapping sequences, enabling rapid imaging of the whole knee for both morphological and quantitative assessment with seamless incorporation into clinical workflow. The overall hypothesis is that a rapid 5-minute knee MRI protocol can be equivalent to the standard 35-minute clinical knee MR protocol. Our proposal includes three specific aims: (i) development of a robust deep learning method for a 4-minute rapid multi-planar morphological knee imaging, (ii) development of a deep learning method for a 1-minute whole-knee-covered high-resolution T2 mapping, and (iii) investigation of a comprehensive evaluation for rapid knee MR protocol in patients with knee osteoarthritis. Successful completion of this project will deliver a rapid 5-minute knee MRI protocol, including routine clinical imaging and additional T2 mapping that can fit into a standard 15-minute clinical time slot. This protocol will be well-evaluated and implemented in clinical settings to facilitate dissemination for further validation. Our methods would offer a unique opportunity to improve joint health care, reduce healthcare costs, and benefit a large population that suffers knee pain and joint discomfort.

项目摘要 普通民众的膝盖疼痛的高流行对公众提出了巨大的挑战 健康，给我们的社会带来巨大的医疗保健和经济负担。磁共振成像（MRI）为 评估膝盖疼痛患者的首选成像方式。确实，外围关节排名第三 最常见的身体部位使用MRI成像，膝盖是迄今为止评估的最常见的关节。给出 随着膝盖发病率的增加，膝盖MRI检查数量的兴起 受伤和膝盖骨关节炎的患病率增加，紧急临床需要减少 膝盖MRI的经济负担，最直接的方法是减少总体时间 进行MRI检查。在过去的十年中，已经尝试了多种技术来加速 膝盖MRI，包括平行成像，压缩感测，多片采集和三维 各向同性分辨率成像。但是，所有当前方法都有局限性，包括降低信号到噪声 比率，图像模糊，不兼容以提出必要的组织对比以及无法评估所有关节 结构。缺乏适当的加速方法还可以防止定量MRI，例如T2松弛时间 尽管在检测早期的关节变性迹象的值中，但在临床上使用的映射。这 应用旨在开发新颖的快速获取和重建技术，以最大化MR Scanner 效率，提高成像管理和自动化扫描工作流程，其最终目标是减少 膝盖MRI的经济负担并促进临床成像操作。我们提出的新方法将是 基于发展高级深度学习重建，并结合新颖的快速图像获取和 自动处理，所有这些都是由我们的研究小组开创的。我们建议开发，优化， 并评估由所有临床序列和其他T2映射组成的快速5分钟膝关节MRI方案 序列，使整个膝盖快速成像，以进行形态学和定量评估 无缝融合到临床工作流程中。总体假设是快速5分钟的膝盖MRI方案 可以等于标准的35分钟临床膝盖MR方案。我们的建议包括三个具体目标： （i）开发一种可用于4分钟快速多平面形态膝关节成像的强大深度学习方法， （ii）开发1分钟全膝盖覆盖的高分辨率T2映射的深度学习方法 （iii）调查膝关节骨关节炎患者的快速膝关节MR方案的全面评估。 该项目的成功完成将提供5分钟的膝盖MRI协议，包括常规临床 成像和额外的T2映射，可以适合标准的15分钟临床时间插槽。该协议将是 在临床环境中进行了充分评估和实施，以促进传播以进一步验证。我们的方法 将提供一个独特的机会来改善联合医疗保健，降低医疗保健成本并受益于大型 患有膝盖疼痛和关节不适的人群。