Machine Learning Analysis for prediction of Osteoporosis using Multimodal Data
Bone fracture due to osteoporosis is an enormous health issue as millions of people over age of 50 around the world are affected. When Bone Mineral Density decreases, the bone becomes porous and fragile. It leads to bone fracture and severe health complications. Research studies have linked various genetic factors to osteoporosis. DXA imaging technique is common and useful to measure BMD and diagnose osteoporosis. Researches are going on to process the images effectively to extract features from the DXA images and use those features to diagnose osteoporosis and osteoporotic fracture. Results alone from DXA scan are not enough to precisely diagnose the disease. There could be various genomic and conventional factors behind osteoporosis. So, we need a robust model that studies genomic, phenotype and image data together to predict osteoporosis. We aim to build a model that will use genotype, phenotype and DXA image data. Using machine learning algorithms on genotype and phenotype data, we can identify people with risk of osteoporosis. The algorithms will generate results which contain genetic factors, phenotype factors behind osteoporosis. The model built by using DXA images of the patients can be used to calculate BMD and predict osteoporosis which will produce efficient results combined with results from genomic analysis. The results will help people get early and proper treatment The research project will continue over a span of 5 years where various machine learning techniques, new algorithms will be tested, and the model will be built and experimented to produce accurate results. Various data pre-processing techniques will be studied and experimented to produce efficient model. Early diagnosis and treatment of osteoporosis can provide early treatment and avoid life-threatening complications for people with low BMD. This research study aims to identify people with risk of osteoporosis. This can save lives as well as save millions of dollars on post osteoporosis treatment.
Osteoporosis is a severe health issue that affects mostly aging people. This disease makes bone porous, weakens the bone strength leading to fractures and life-threating complications. Millions of people are affected worldwide. There are advanced assessment tools like Dual-energy X-ray Absorptiometry (DXA), which is used to measure Bone Mineral Density (BMD). Low BMD is one of the reasons which leads to osteoporosis. There are various genotypes, phenotypes factors that cause this disease. People can prevent osteoporotic fractures by changing their lifestyle if osteoporosis and osteoporotic fracture can be detected or predicted early.
Researchers are using machine learning algorithms to analyze the genomic and phenotype data and predict the osteoporosis. Various algorithms on DXA images are used to calculate BMD and forecast osteoporosis. These researches have produced useful results, but there is little research on study of genomic data and DXA image data together.
We aim to build a robust model which analyzes the DXA images of various part of the body using Deep Learning to predict the osteoporosis and use the Machine learning techniques to analyze the genotype, phenotype data which will discover the mechanism of osteoporosis and the model can be used to predict the osteoporosis.
Previous scope extension
The human skeletal system consists of bones, joints, cartilage, tendons, and ligaments in the body. Despite their hardness and strength, bones and joints can suffer from injury and disease, including osteoporosis, osteoporotic fracture, osteoarthritis (OA), rheumatoid arthritis (RA), and others. Millions of people are affected by skeletal diseases worldwide, so early identification of high-risk individuals would be significant, which will help diminish the duration and severity of skeletal illnesses and ultimately reduce the burden on society. However, limitations exist among current predictive models for mentioned diseases, e.g., FRAX, the most commonly used fracture prediction tool, does not include genetic information in the predictive algorithm. We aim to use novel methods to build robust models for predicting osteoporosis, osteoporotic fracture, OA, and RA, respectively.
These conditions mentioned above, especially osteoporosis and RA, have high comorbidity with psychiatric and neurological disorders, which include schizophrenia, bipolar disorder, major depression, attention-deficit hyperactivity disorder, autism spectrum disorder, and Alzheimer's disease. Medication and unhealthy lifestyle behaviors may contribute to the comorbidity, but shared genetic risk factors and environmental factors may also play a role. Recently, substantial evidence has accumulated to indicate that in many psychiatric disorders, there are gene variants related to abnormalities in phospholipid metabolism, which may also be the case in osteoporosis. Nevertheless, the genetic relationship between the two common disease domains is under-studied. Besides, causes of human diseases are multifactorial including shared genetic and environmental risks as well as the interplay of genes and the environment. It has become crucial to incorporate gene-environment interactions (GxE) in the study of complex traits. We will also use the UKB data to propose a statistical test for detecting interaction effects of an environmental factor and a set of genetic markers containing both rare and common variants.