The influence of the normalisation of spinal CT images on the significance of textural features in the identification of defects in the spongy tissue structure

The aim of the study was to determine the effect of normalisation of spinal CT images on the accuracy of automatic recognition of defects in the spongy tissue structure of the vertebrae on the thoraco-lumbar region. Feature descriptors were based on the grey-levels histogram, gradient matrix, run-length matrix, coocurrence matrix, autoregression model and wavelet transform. Six methods of feature selection were used: Fisher coefficient, minimisation of classification error probability and average correlation coefficients between chosen features, mutual information, Spearman correlation, heuristic identification of noisy variables, linear stepwise regression. Selection results were used to build 6 popular classifiers. The following values of individual classification quality factors were obtained (before normalisation/after normalisation): general accuracy of classification - 90%/82%, classification sensitivity - 89%/85%, classification specificity - 96%/82%, positive predictive value - 95%/95%, negative predictive value - 89%/84%. For the applied set of textural features, as well as the methods of selection and classification, image normalisation significantly worsened the accuracy of the automatic diagnosis of osteoporosis based on CT images of the spine. Therefore, it is necessary to use this operation with caution so as not to remove from the processed images information significant from the point of view of the purpose of the research.