AI-based Diagnosis And Prediction Of Biochemical Testing For Osteoporosis

Osteoporosis has become a public health problem in the world due to its aggressive hazard to human beings.Fragility fractures caused by osteoporosis can result in disability and even death,which seriously affect the life quality of patients,and put great and long-term economic pressure on patients,families and society.Osteoporosis comes with aging and endocrine changes,which are difficult to be noticed in human daily life.Most of people go to the doctor after fracture happening,and the disease has become more serious at this time.Therefore,the essential and the only way to deal with the threat of osteoporosis is to achieve early identification,early prevention and easily monitoring through simple daily examination.Studies have found that the bone metabolism disorder is the basic pathophysiological mechanism of various bone metabolic diseases including osteoporosis.Biochemical test of bone turnover markers can be used as routine examination in hospitals,which has the potential to realize above purpose of diagnosing osteoporosis in the general population.However,as a degenerative disease with a complex metabolism network,it is difficult to find markers with high specificity and sensitivity.In order to obtain an accurate diagnosis,it is necessary to detect multiple markers,which is costly and difficult to operate.Even so,the diagnosis accuracy of osteoporosis based on bone metabolism cannot be guaranteed even by experienced doctors.To sum up,based on clinical bone metabolism test and combined with artificial intelligence theory,a rapid,noninvasive,economic and reliable method of diagnosing osteoporosis is explored.In study on the diagnosis of primary osteoporosis in this paper,several bone-related biochemical indices from91 female patients and 50 normal female are collected,and a variety of machine learning model are built,such as kNN,decision tree,random forests,gauss bayesian,support vector machine and neural network.The goal of primary osteoporosis screening based on age,Ca,P,ALP,CTx and VD was finally achieved.And the SVM model gets the best classification performance with an accuracy of 82%,and a recall rate of 82%.In the study of secondary osteoporosis diagnosis,several bone-related biochemical indices from 40 patients of type 2 diabetic osteoporosis and 162 patients of type 2 diabetes are collected,and a variety of machine learning model are built.The goal of primary osteoporosis screening based on sex,age,BMI,Ca,P,ALP,TPINP and PICP was finally achieved.And the SVM model gets the best classification performance with an accuracy of 87%,and a recall rate of 90%.For exploring the new target of diagnosing and predicting osteoporosis,enhanced raman spectroscopy detection of clinical urine samples are implemented.It was found that the urine Raman spectroscopy data has a positive performance in the biclassification task of normal bone mass and decreased bone mass based on BP neural network modeling,and the classification accuracy reached 75%,which suggests that urine Raman spectroscopy may be a new target for the prediction and diagnosis of osteoporosis.The study shows that the diagnostic method of combining biochemical test with AI,instead of originalclinical destructive detection is feasible and might have advantages.Hence,AI plays a role in promoting clinical disease diagnosis.In the future,with more comprehensive data and further researches,it may be possible to change the golden standard completely,and develop a new non-invasive,efficient,widespread and economic method for diagnosing osteoporosis.