Abstract:
Osteoporosis, a disorder defined by decreased bone mineral content and changes in bone microarchitecture, poses a
challenge for accurate classification using X-ray images. This work aims to extract texture features from calcaneal radiographs and
select the best texture features which can be used to train the machine learning classifier models for detection of osteoporosis. This
work is based on multiresolution analysis and microstructural analysis to characterize trabecular bone microarchitecture from
calcaneal radiograph. The image is transformed to extract the feature details using a two-level wavelet decomposition. Structural
texture methods such as Local Binary Pattern, fractal dimension and Gabor filter are applied to the wavelet decomposed images. The
most discriminating texture features are selected using independent sample t-test and feature selection methods. Machine learning
models are constructed by training the classifiers using the best texture features to classify healthy images from osteoporotic images.
The effectiveness of the proposed approach is evaluated using a public challenge dataset comprising calcaneal radiographic images.
Notably, the best classification is obtained with k-Nearest Neighbour trained with the features selected using forward feature
selection, with an accuracy rate of 78.24%. The results indicate the potential of the proposed approach as a possible alternative tool
for screening osteoporosis.
