Image processing of radiographs in the assessment of osteoporosis.

On radiographs cancellous (trabecular) bone structure appears as a distinct pattern. Osteoporosis results in characteristic changes in this pattern, and distinctive difference in appearance of healthy and osteoporotic bone. The usefulness of the pattern change for the diagnosis of osteoporosis was first recognised in the 1960s using radiographs of the proximal femur (i.e. the Singh Index grading system) but has always been considered too variable for diagnosis or epidemiology. More recently image processing techniques have shown promise as a methodology which can overcome the limitations of the observer grading. Although there is a growing body of evidence that such techniques may indeed be useful, the studies are far from conclusive or near to providing a tool for the study of osteoporosis. The aim of this doctoral work was to further the study of the changes in trabecular pattern caused by osteoporosis, and identify ways in which the pattern changes, using established image processing techniques. The work examined a number of digital image processing techniques from the "Texture Analysis" paradigm as potential tools for the study of trabecular patterns. Radiographs and clinical data from 140 women (aged between 44 and 66 years) enrolled in a large London District General Hospital study were used in the principal study. Femoral neck trabecular pattern "texture" was correlated with fracture risk factors: bone densitometry, body mass, age and the visual grading of trabecular hip patterns (Singh Index). The changes in trabecular texture caused by the menopause, the symmetry of left and right limbs, and fracture rates were also explored. Research was also undertaken to explore and characterise sources of error and to demonstrate reproducibility. This research has found that there is a statistical relationship between densitometry, age, body mass, Singh Index and differences of patients between pre- and post- menopause for certain textural properties. More significantly, the texture analysis method can predict fractures indicative of osteoporosis. Reproducibility was established, and error sources isolated. This thesis proposes that image processing technology can be used for assessing osteoporotic bone changes in epidemiological and populational studies, and can support bone densitometry data at little extra cost. Moreover, the texture analysis of the proximal femur can be used as an osteoporotic fracture risk factor in its own right and could be developed further as clinical diagnosis modality. The system used in these studies has a low signal-to-noise ratio, however, and is not suited to diagnostic situations at present. Moreover, this thesis does not establish a physical link between radiographic trabecular texture and osteoporotic changes of cancellous structural, and the link is statistical in nature. Further study would be required to establish the technology on a firmer footing.