A population-based study of deep learning derived organ and tissue measures for accelerated aging using repurposed abdominal CT images

项目总结 生活在身体机能减退和精神障碍的人数急剧增加。 与衰老相关的慢性疾病。如果我们将时序年龄(基于日历的年龄)与生物年龄进行比较 (细胞、组织、器官和系统水平的变化)，我们可以更快地将人归类为老龄化 (加速衰老)或慢于同龄人(成功衰老)。已经开发出了测量方法 基于DNA甲基化、端粒长度和血液生物标志物的生物年龄。然而，这些措施 可能不能准确反映衰老引起的器官和组织水平的变化。一种多器官/组织的方法是 需要在出现体征、症状或临床症状之前确定与年龄相关的全面结构变化 诊断就会发生。腹部计算机断层扫描(CT)在普通人群中得到了广泛应用(35% 年龄在20-89岁的成年人在11年内)。人体器官和组织的定量测量 腹部CT可以预测特定器官的疾病，或者结合使用可以用来计算生物年龄 并预测住院和死亡率的更具全球性的结果。因此，我们的中心假设是 深度学习模型应用于腹部CT可以量化器官和组织的结构特征 识别衰老加速的人，器官特定疾病、住院和死亡的高危人群。 罗切斯特流行病学项目记录链接系统提供了一个通用的人口档案 423,081次腹部CT的图像和181,187名成年人(年龄)的综合病历数据 20-89岁)2010-2020年间。我们的团队已经开发并验证了测量肝脏的DL工具， 腹部CT图像上的肾脏、主动脉、脂肪、肌肉和骨骼。我们将利用这些资源1)建立 来自健康和普通人群样本的腹部CT生物标记物的百分位数；2)确定 腹部CT生物标志物在普通人群中患器官特异性临床疾病的风险；以及3)确定 一般人群中与腹部CT措施相关的住院和死亡风险。如果 DL工具在腹部CT图像中的成功应用将丰富与年龄相关的表征 健康风险，无需额外的检测负担。亚临床腹部CT生物标记物也可能为 研究衰老和早期疾病的生物学，改进疾病分类，为早期提供机会 干预。

PROJECT SUMMARY There has been a dramatic increase in the number of persons living with reduced physical function and with aging-related chronic conditions. If we compare chronological age (calendar-based age) with biological age (changes at the cellular, tissue, organ, and system levels), we can classify persons as aging faster (accelerated aging) or slower (successful aging) than their peers. Methods have been developed to measure biological age based on DNA methylation, telomere length, and blood biomarkers. However, such measures may not accurately reflect organ- and tissue-level changes from aging. A multi-organ/tissue approach is needed to identify comprehensive age-related structural changes before signs, symptoms, or clinical diagnoses occur. Abdominal computed tomography (CT) has widespread use in the general population (35% of adults ages 20-89 years in an 11-year period). Quantitative measures of the organs and tissues on abdominal CT may predict organ-specific diseases, or in combination, may be used to calculate biological age and predict the more global outcomes of hospitalization and mortality. Therefore, our central hypothesis is that deep learning (DL) models applied to abdominal CTs can quantify structural features of the organs and tissues to identify persons with accelerated aging at high-risk for organ-specific disease, hospitalization, and death. The Rochester Epidemiology Project record-linkage system provides access to a general population archive of images for 423,081 abdominal CTs and to comprehensive medical record data among 181,187 adults (ages 20-89 years) between 2010-2020. Our team has already developed and validated DL tools to measure liver, kidney, aorta, fat, muscle, and bone on abdominal CT images. We will leverage these resources to 1) establish percentiles of abdominal CT biomarkers from both healthy and general population samples; 2) determine the risk of organ-specific clinical disease by abdominal CT biomarkers in the general population; and 3) determine the risk of hospitalization and death associated with abdominal CT measures in the general population. If successful, application of DL tools to abdominal CT images will enrich the characterization of age-related health risks without additional testing burden. Subclinical abdominal CT biomarkers may also inform the biology of aging and early disease, improve disease classification, and provide opportunities for early intervention.