Application of wavelet analysis to ultrasonic characterization of bone

Investigates the application of time-frequency techniques to ultrasonic signals transmitted through bone to noninvasively determine bone density and strength. Ultrasonic transmission measurements are carried out on 3 samples of human vertebral trabecular bone, using a 1 MHz transducer pair. The data is processed using the wavelet transform according to Mallat's (1989) pyramid algorithm. The wavelet representation thus obtained is then cross-correlated with a wavelet representation of a reference signal to obtain the time-frequency representation associated with a particular bone sample. Results indicate that time-frequency methods may serve to accurately discriminate bone in various biophysical states. Further studies are under way to determine the best set of features associated with the time-frequency representation as well as the optimal means for processing the feature set, including neural networks.<>