An intelligent approach to predict thermal injuries during orthopaedic bone drilling using machine learning

The thermal injuries increase with temperature elevation during bone drilling that can cause irreversible, permanent death of regenerative bony cells resulting in thermal osteonecrosis. The ascent in temperature during the drilling procedure is a significant concern for every orthopaedic surgeon. Since it is difficult to monitor and predict temperature elevation during real-time in vivo medical surgery, a robust predictive machine-learning (ML) model has been proposed in the present work. Successively, the efficiency of rotary ultrasonic-assisted bone drilling (RUABD) is experimentally verified to reduce thermal injuries during bone drilling. Several rigorous in vitro experiments were performed on pig femur bone with changing independent variables like rotational speed, feedrate, abrasive grit size, and vibrational ultrasonic power during the study. The assumptions for the implementation of machine learning models have been successfully corroborated and validated. The multi-linear regression was compared with a multilayer perceptron, lasso regression, and ridge regression to provide the most accurate predictive models. The accuracy of ML models was observed with different error metrics such as mean absolute error (MAE), root mean square error (RMSE), and mean square error (MSE). The error metrics of ridge regression were comparatively lower with MAE 1.702 +/- 0.229, RMSE 2.015 +/- 0.398 and MSE 5.214 +/- 1.840 than other ML models. The Ridge regression model was able to predict temperature rise during bone drilling with an adequacy of +/- 1.7 degrees C. The prediction of thermal injuries using machine learning models is the key contribution and a proof-of-concept of the present in vitro study.

Automated summary

A machine-learning model has been proposed in this study to predict temperature elevation during bone drilling and reduce thermal injuries. Through in vitro experiments, the accuracy of the machine-learning models was validated, and the ridge regression model was found to be the most accurate predictive model.