Predicting Osteoporosis with Various Machine Learning Algorithms using Dual-energy X-ray Absorptiometry: a comparative analysis

Abstract

Purpose: Osteoporosis is a condition where bone density decreases, impacting bone quality and increasing susceptibility to fractures. Diagnosis typically involves imaging techniques like DEXA scan. To enhance early detection, new predictive algorithms are essential due to limitations in clinical diagnostic software. This study aims to predict osteoporosis with various machine learning (ML) algorithms and pinpoint factors contributing to the disease based on DEXA scans of the femoral neck and lumbar spine.

Methods: In this study, we analyzed the data from 1000 people who were encountered for densitometry at the Rheumatology Clinic in a major metropolitan general hospital for predicting osteoporosis, we used five classification algorithms, including Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Logistic Regression (LR). We evaluated the performance of models using four metrics, including accuracy, sensitivity, specificity, and Area Under the Receiver Operating characteristic Curve (AUROC). All of programing was done using the Python programming language in the Google Colab environment.

Results: Out of 1000 patient records, there were 761 women and 239 men, with a 58.42 mean age. Osteoporosis occurred in 23.5% of cases. ANN and RF, with 89% and 78.6%, had the highest sensitivity, respectively. ANN and SVM, with 96.3% and 94.2%, had the highest specificity. In accuracy, ANN and RF, with 94.6% and 86.5%, were the highest. Based on the AUROC, the ANN method achieved the best performance (0.937), followed by RF (0.837), LR (0.832), SVM (0.769), and DT (0.715).

Conclusion: The ANN model emerged as the strongest performer, achieving high sensitivity, specificity, and overall accuracy. The study’s findings hold promise for enhancing the earlier diagnosis of osteoporosis. Machine learning algorithms can provide an alternative approach to identifying and screening individuals at high risk for osteoporosis and can be used in the development of clinical decision support systems for the diagnosis of osteoporosis.