Bone fracture risk prediction based on clinical variables and DXA results by support vector machine

Background and aims: Osteoporosis is defined as bone strength impairment due to mineral depletion,which puts the patient at risk of fracture. Quantitative assessment of bone mineral density(BMD) is performed by dual energy X-ray absorptiometry (DXA). Then the obtained results arejudged according to the T-score index. The T score, together with information related to a person’shealth and lifestyle, is one of the important components for calculating fracture risk, which is usedin fracture risk prediction calculators such as CAROC and FRAX.In order to prevent fractures caused by osteoporosis, it is necessary to assess the risk of fracture.Some of the studies that have been done before in this field have proposed methods for diagnosingosteoporosis and determining bone density based on the processing of radiographic images,but radiographic images are only two-dimensional images that give us less information about thetexture and porosity of the bone. they give. Therefore, using DXA method can be more beneficialin order to get the exact amount of bone density. The proposed methods often have high computingtime and are not easy to use for diagnostic predictions for the general public. To facilitate theprediction of osteoporosis with more available data, the use of machine learning methods can bebeneficial.The purpose of this study is to use a machine learning model based on neural networks that canpredict the risk of fracture based on the specifications of the patient’s clinical and laboratory variablesand DXA results. Another goal is that the proposed method can be used by any type of userwith high accuracy and speed.Method: The dataset used in this article was collected from ۸۱۷ person in the age range of ۵۰ to۹۸ years. The features in the dataset include the history of diseases, effective behaviors, the typeof fracture and the level of bone density, etc. Based on the features, several supervised learningalgorithms have been used to predict the possibility of fracture occurrence in future years. The finalproposed method is based on neural network and support vector machine algorithm. The finalmodel will be evaluated based on the cross-validation method, confusion matrix, etc.Results: After pre-processing and data standardization, several models of supervised learningalgorithms including perceptron neural network, support vector machine, k nearest neighbor, decisiontree were used. The SVM neural network obtained the highest accuracy among the algorithms,which could be used for the purpose of fracture risk prediction with ۹۴.۴۸% accuracy.Conclusion: Neural networks can be used as a suitable model for osteoporosis classification byfinding the best possible boundary between two classes of data. SVM works very well on high-dimensionaldata and is well suited to binary classification problems. According to the obtained results,the use of support vector machine can be beneficial in predicting the probability of fracturein the coming years.

Keywords:

Osteoporosis , machine learning , SVM , BMD , Fracture risk prediction