Developing an Ultrasonic Approach for Bone Density Assessment Via Diffracted Pulse Strength Analysis
An ultrasonic bone density measurement is a radiation-free alternative to conventional methods like dual-energy X-ray absorptiometry (DXA). This research investigates the feasibility of bone density estimation from features of ultrasound signals using machine learning. A 1000-sample dataset characterized by features like bone region, wave speed, signal amplitude, frequency, time-of-flight, noise, and signal-to-noise ratio (SNR) was analyzed. Feature engineering was employed to derive composite features from raw signal features. A training MLP regressor with the engineered feature set provided a promising R-squared of 0.6723, with a mean absolute error of 0.1426 g/cm³. The findings show that ultrasonic signal analysis with the appropriate machine learning methodologies has excellent prospects in non-invasive bone density estimation. This approach could ultimately result in safer and more accessible methods for tracking bone health, particularly in populations where radiation exposure is a concern. Future steps would be to validate this process with in-vivo data and refine the feature engineering process to further improve predictive accuracy.