A Systematic approach to improve Support Vector Machine applied to ultrasonic guided wave spectrum image classification

Osteoporosis is a skeletal disorder characterized by low bone mass, which compromises its resistance and increases the risk of fractures, and is a widespread problem worldwide. Currently, the gold standard for assessing fracture risk is the measurement of the areal bone mineral density with Dual-Energy X-ray Absorptiometry. Several ultrasound techniques have been presented as alternatives. It has been shown that the estimation of cortical thickness and porosity, obtained by Bi-Directional Axial Transmission, are associated with non-traumatic fractures in postmenopausal women. Cortical parameters were derived from the comparison between experimental and theoretical guided modes. However, this model-based inverse approach tends to fail for the patients associated with poor guided mode information. A recent study has shown the potential of an automatic classification tool, Support Vector Machine, to analyze guided wave spectrum images independently of any waveguide model. The aim of this study is to explore how the classification accuracy varies with the number of features. Optimization was done using the Particle Swarm Optimization algorithm, while adjustment was made considering age, body mass index, and cortisone intake. The results show that adjusting the data and optimizing the parameters improved classification. Moreover, the number of features was reduced from 32 to 15, with 73.5% accuracy comparable to the gold standard.