Study Of Photoacoustic Imaging In The Diagnosis Of Osteoporosis

Osteoporosis is a common systemic metabolic bone disease,which is characterized by the decreasing in bone mass as well as the changing in bone micro-structure.This disease is more common in postmenopausal women and older men.The serious consequence of osteoporosis is the occurrence of osteoporotic fractures that can occur during minor trauma or daily activities.In recent years,the incidence of osteoporosis in China has increased so much and it has become a hidden danger to the health of the elderly in China.Moreover,the age of the osteoporosis patients has the tendency to be younger,so it is important to find some quick and effective methods for testing osteoporosis.This article is mainly to explore some new methods in order to diagnose osteoporosis,including traditional qualitative spectrum analysis and new quantitative machine learning methods.In recent years,the Quantitative Ultrasound Spectroscopy(QUS)is an emerging method for detecting osteoporosis.The degree of the changing in bone is determined mainly by the trend of the slope of spectrum.We first demonstrate the feasibility of mathematical derivation.Then we compare the trend of photoacoustic spectrum with the broadband ultrasound spectrum method to show that the photoacoustic spectrum method can equally detect osteoporosis.Both the simulation experiment and the actual experiment prove that the two have the same changing trend.We determine the trend of photoacoustic spectrum change according to the degree of corrosion of osteoporosis,and then according to the three groups of bone conditions,the bone signal photoacoustic spectrum and the ultrasound spectrum comparison reached the final conclusion.In addition to qualitative methods,this paper also studies quantitative methods for detecting osteoporosis.We mainly use k-means clustering,support vector machine,and convolutional neural network.For the first two methods,we extracted a dozen or more feature vectors including peak,mean,and spectral slope to train as parameters,while CNN classifies the signal as a one-dimensional image,and finally svm and CNN achieved a better effect of classification.