Deep Learning Analysis of Ultrasonic Guided Waves for Cortical Bone Characterization

Area of Science:

• Biomedical Engineering
• Materials Science
• Artificial Intelligence in Healthcare

Background:

• Characterizing long cortical bone using ultrasonic guided waves (UGWs) presents a complex multiparameter inverse problem.
• Traditional inverse problem solutions often involve computationally intensive optimization processes.
• Deep neural networks (DNNs) offer a powerful alternative for predicting multiple parameters by learning complex mapping relationships.

Purpose of the Study:

• To investigate the feasibility of applying a multichannel crossed convolutional neural network (MCC-CNN) for simultaneous estimation of cortical thickness and bulk velocities (longitudinal and transverse) in long cortical bone.
• To develop a method that avoids direct multiparameter optimization problems inherent in UGW-based bone characterization.
• To validate the MCC-CNN's predictive accuracy using both simulated and experimental UGW data.

Main Methods:

• Utilized the finite-difference time-domain (FDTD) method to generate simulated UGW array signals for training the MCC-CNN.
• Developed and trained a multichannel crossed convolutional neural network (MCC-CNN) to establish a mapping between UGW signals and cortical bone material parameters.
• Validated the trained network on simulated data, bone-mimicking phantom experimental data, and ex vivo bovine cortical bone data.

Main Results:

• The MCC-CNN accurately predicted longitudinal bulk velocity (V_L), transverse bulk velocity (V_T), and cortical thickness (Th) with low root-mean-squared errors (RMSE) on simulated data (97 m/s, 53 m/s, 0.089 mm, respectively).
• Experimental validation on bone-mimicking phantoms yielded comparable RMSE values (120 m/s, 80 m/s, 0.14 mm for V_L||, V_T||, and Th, respectively).
• Predicted parameters from ex vivo bovine cortical bone experiments showed good agreement with theoretical dispersion curves.

Conclusions:

• The proposed MCC-CNN method provides a feasible and accurate approach for evaluating long cortical bone properties using UGWs.
• This deep learning technique effectively bypasses complex multiparameter optimization, offering a more streamlined inverse problem solution.
• The study demonstrates the potential of MCC-CNN for non-invasive, quantitative assessment of bone health and material characteristics.