Application of texture analysis for automated osteoporosis diagnostics by plain hip radiography

The trabecular bone microarchitecture is used by radiologists for visual bone diseases diagnostics. So the possibility of the automated computer diagnostics of the digital radiograms was studied. A number of textural features were examined to select the most efficient group for the diagnostic. Plain hip radiograms of 42 patients were used for both BMD measure and textural features evaluation. For the features efficiency assessment both the variance analysis and the correlation analysis were used. Also the efficiency was validated experimentally by the diagnostic error probability assessment. It was found that the most efficient features are the covariance features—they provide diagnostic error probability 0.2. Also there was a significant correlation (0.6) between some Gabor spectral features and the Singh Index performed by the radiologists.