Quantitative MRI and Deep Learning Technologies for Classification of NAFLD

用于 NAFLD 分类的定量 MRI 和深度学习技术

• 批准号: 10668430
• 负责人: Maria I. Altbach
• 金额: $ 57.49万
• 依托单位: UNIVERSITY OF CALIFORNIA LOS ANGELES
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-08-01 至 2027-04-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10668430
• 关键词: Abdomen; Acceleration; Adoption; Area; Arizona; Attention; Biological Markers; Biomedical Engineering; Biopsy; Breathing; California; Cells; Cessation of life; Characteristics; Cirrhosis; Classification; Clinic; Clinical; Compensation; Data; Detection; Development; Diagnosis; Disease; Early Diagnosis; Fatty Liver; Fatty acid glycerol esters; Fibrosis; Future; Goals; Health Care Costs; Hepatic; Image Enhancement; Inflammation; Interdisciplinary Study; Interobserver Variability; Iron Overload; Joints; Label; Life; Liver; Liver Failure; Liver Fibrosis; Liver parenchyma; Los Angeles; MRI Scans; Magnetic Resonance; Magnetic Resonance Elastography; Magnetic Resonance Imaging; Maps; Measures; Medical Imaging; Modeling; Monitor; Morbidity - disease rate; Morphologic artifacts; Motion; Outcome; Patients; Performance; Perfusion; Physics; Primary carcinoma of the liver cells; Protocols documentation; Protons; ROC Curve; Radial; Reproducibility; Research; Risk Factors; Sampling Errors; Scanning; Scheme; Signal Transduction; Staging; Techniques; Technology; Testing; Time; Tissues; Training; Uncertainty; Universities; Validation; Water; chronic liver disease; contrast enhanced; deep learning; deep learning model; density; disease classification; elastography; hepatocyte injury; image processing; improved; learning strategy; liver biopsy; liver inflammation; magnetic resonance imaging biomarker; new technology; non-alcoholic fatty liver; non-alcoholic fatty liver disease; nonalcoholic steatohepatitis; novel therapeutics; prospective; prospective test; reconstruction; sensor; simple steatosis; treatment response

PROJECT SUMMARY Non-alcoholic fatty liver disease (NAFLD) is the most common chronic liver disease in the U.S. and ranges from simple fatty liver (or non-alcoholic fatty liver, NAFL) to the progressive form, non-alcoholic steatohepatitis (NASH). About 20-30% of subjects with NAFL develop NASH, which is caused by hepatocyte injury, hepatic inflammation, and resultant hepatic fibrosis. NASH can lead to life-threatening conditions, but is difficult to diagnose at early stages. Liver biopsy is the current standard to diagnose NAFL/NASH, but biopsy is invasive, has associated morbidity, and is limited by sampling errors and inter-observer variability. Many patients present with later stage NASH, adversely impacting outcomes and healthcare costs, which are estimated at $32 billion annually in the U.S. Magnetic resonance imaging (MRI), including elastography (MRE), is a technology that can non-invasively quantify hepatic fat (MRI proton-density fat fraction), iron overload (MRI R2*), and fibrosis (MRE stiffness). However, current liver MRI is challenged by motion artifacts and incomplete signal models, which can compromise the accuracy and reproducibility of the quantitative parameters derived from them. In addition, early tissue changes associated with NASH are not adequately characterized using conventional MRI. The common requirements of breath-holding and long protocols also severely limit the adoption of liver MRI in the clinic. Furthermore, the present clinical interpretation of MRI has limited ability to distinguish NASH from NAFL. The research teams at the University of California Los Angeles, University of Arizona, and Siemens have been leading the development of motion-robust radial MRI to quantify hepatic PDFF and R2*, T2 and T1, perfusion, and stiffness. The Siemens team has also developed deep learning methods for medical image processing and disease detection and classification. In this bioengineering research partnership project, the multi-disciplinary research team will investigate four aims: (1) Develop a robust motion compensation framework for free-breathing multi-parametric quantitative radial liver MRI; (2) Accelerate quantitative liver MRI scans through combined acquisition and joint modeling of multiple parameters, data undersampling, and deep learning-based reconstruction and quantification; (3) Develop deep learning models to accurately classify NAFL versus NASH and measure the degree of fibrosis based on quantitative MRI; (4) Prospectively assess the new quantitative MRI and deep learning technologies for classifying NAFL versus NASH and measuring fibrosis in patients, with respect to liver biopsy. The new free- breathing quantitative MRI and deep learning technologies developed in this project will accurately classify NAFL versus NASH and measure fibrosis using data from the entire liver and thus help to avoid liver biopsy, allow monitoring of treatment responses, and accelerate the development and implementation of new therapies.

项目概要 非酒精性脂肪肝 (NAFLD) 是美国最常见的慢性肝病，范围广泛 从单纯性脂肪肝（或非酒精性脂肪肝，NAFL）到进行性非酒精性脂肪性肝炎 （纳什）。大约 20-30% 的 NAFL 受试者会发展为 NASH，这是由肝细胞损伤、肝 炎症，以及由此引起的肝纤维化。 NASH 可导致危及生命的疾病，但很难治愈 早期诊断。肝活检是目前诊断NAFL/NASH的标准，但活检是侵入性的， 具有相关的发病率，并受到抽样误差和观察者间变异的限制。很多病人 出现晚期 NASH，对结果和医疗费用产生不利影响，估计为 美国每年 320 亿美元的磁共振成像 (MRI)，包括弹性成像 (MRE)，是一项 可以非侵入性量化肝脂肪（MRI 质子密度脂肪分数）、铁过载（MRI R2*) 和纤维化（MRE 硬度）。然而，当前的肝脏 MRI 受到运动伪影和不完整的挑战 信号模型，这可能会影响所导出的定量参数的准确性和再现性 来自他们。此外，与 NASH 相关的早期组织变化尚未通过以下方法得到充分表征： 常规 MRI。屏气和长时间协议的共同要求也严重限制了 临床上采用肝脏MRI。此外，目前 MRI 的临床解释能力有限 区分 NASH 和 NAFL。加州大学洛杉矶分校、加州大学洛杉矶分校的研究团队 亚利桑那州和西门子一直在引领运动稳健的径向 MRI 的开发，以量化肝脏 PDFF 和 R2*、T2 和 T1、灌注和硬度。西门子团队也开发了深度学习 医学图像处理以及疾病检测和分类的方法。在这个生物工程 研究伙伴关系项目中，多学科研究团队将研究四个目标：（1）开发 用于自由呼吸多参数定量径向肝脏 MRI 的稳健运动补偿框架； (2) 通过多个数据的组合采集和联合建模加速定量肝脏 MRI 扫描 参数、数据欠采样以及基于深度学习的重建和量化； （三）深层次发展 学习模型可准确分类 NAFL 与 NASH 并根据以下指标测量纤维化程度 定量磁共振成像； (4) 前瞻性评估新的定量MRI和深度学习技术 根据肝活检对 NAFL 与 NASH 进行分类并测量患者的纤维化情况。新的免费—— 该项目开发的呼吸定量MRI和深度学习技术将准确分类 NAFL 与 NASH 相比，并使用整个肝脏的数据来测量纤维化，从而有助于避免肝活检， 允许监测治疗反应，并加速新疗法的开发和实施 疗法。