Image Analysis and Machine Learning Methods for Biomarkers of Age-related and Metabolic Diseases

点击翻译按钮获取中文摘要

Abstract Age-related and metabolic diseases such as type-2 diabetes, cardiovascular diseases, and sarcopenia have become a worldwide epidemic that affects the quality of life of millions. To give a global perspective, roughly 343.8 million people in the world have type-2 diabetes today, and 175 million do not know they have diabetes at all. Metabolic diseases, such as diabetes and osteoporosis, are strongly linked to longitudinal changes in body composition, morphology and function. Modern medical imaging technologies offer the opportunity to study the composition and morphometry of human body in ways that were previously impossible. Contemporary imaging studies that are performed in vivo on a large number of participants have enabled cross-sectional and longitudinal studies of age-related and metabolic diseases, and effects of pharmacological interventions. The emergence of advanced imaging technologies has also created the need for automated image analysis techniques for identification and quantification of morphological patterns of anatomies and tissues and their changes with increasing age. This project will contribute novel and non-invasive medical image analysis techniques for studying the human body composition to achieve timely prognosis of these pathologies. Our research interests will concentrate on identification of morphological patterns in the mid-thigh, abdomen and lower leg that will eventually lead to development of imaging biomarkers. The accumulation of adipose tissue in the human body and changes of its regional distribution are associated with type-2 diabetes, cardiovascular diseases and the metabolic syndrome. Age-related changes in skeletal muscle composition are strongly linked to loss in muscle strength and mass, frequently termed as sarcopenia, leading to decreased mobility and function. Also, trabecular bone structural changes are associated with osteoporosis. We will use imaging and clinical data collected by the Baltimore Longitudinal Study of Aging (BLSA) that is the longest ongoing epidemiology study in the US. This work will address a technical and a clinical hypothesis. The technical hypothesis is that quantitative image analysis can accurately and robustly segment, register and fuse body composition data acquired by modern MRI and CT imaging scanners. The clinical hypothesis is that qualitative body composition phenotypes on clinical imaging can be used as biomarkers for prognosis and diagnosis of the metabolic syndrome manifestations. We will build on recent advances in medical image analysis to contribute novel and non-invasive techniques for studying the human body composition and its longitudinal changes with main applications in tissue identification and quantification at the mid-thigh, lower leg and the abdomen (aim 1). Then we will develop statistical machine learning methods to achieve timely diagnosis and prognosis of metabolic and age-related conditions including the metabolic syndrome and osteoporosis, and to track the effect of interventions (aim 2).