Establishing correlation between bone mineral apposition rate and physiological loading using machine learning regressor

Ageing related post-menopausal osteoporosis is a global healthcare problem for orthopaedic clinicians. As age creeps in, the body tends to have an inhibition in the rate of estrogen production that negatively reflects on the bone mineral apposition rate (BMAR). The present study attempts to establish a correlation between the loading parameters such as strain magnitude, frequency, and the number of cycles on BMAR using Machine learning regressors namely Random Forest Regressor, Support Vector Machine Regressor, K-Nearest Neighbours Regressor, and XGBoost Regressor. The present model is trained and validated using the experimental data, which is later used to simulate the feature importance of loading parameters. XGBoost Regressor outperforms during the prediction of BMAR at the periosteal and endosteal surface. A strong correlation of R² = 0.945 and 0.98 and mean squared error (MSE) of 0.004 and 0.007 was observed on periosteal and endosteal surfaces, respectively. In addition, Quantile-Quantile (Q-Q) plot showed that the endosteal surface have greater potential than the periosteal surface in accurately estimating BMAR and through the feature importance graphs the frequency is observed to be the most significant factor in BMAR on both surfaces. The conclusion of the present study indicates that the XGBoost Regressor has the highest accuracy while predicting BMAR, while the endosteal surface shows greater potential than the periosteal surface in accurately estimated BMAR.