Shape and texture measurements in clinical images of the hip in osteoporosis and osteoarthritis
The aim of this project was to develop and apply image analysis methods to assess the shape and trabecular structure of the femur and determine their relationship with osteoporotic hip fracture. Changes in bone shape and structure caused by osteoarthritis (OA) were also examined. As osteoporosis (OP) progresses, bone is lost, the skeleton becomes progressively weaker and ultimately fractures. Fracture risk is generally assessed clinically by bone mineral density (BMD), a measure of bone quantity. However, the shape and trabecular structure of the bone are also important, but less well-characterised, factors This study applied quantitative measures of shape and structure to medical images of both high (histological sections) and lower resolution (radiographs). Trabecular structure was analysed using the Fourier transform and was tested first on histological sections. Principal component, discriminant and neural network analyses were used to generate classifiers that distinguish between control, OP-fracture and OA subjects. A neural network achieved 84% accuracy, compared with 70% for the best discriminant analysis. Femoral morphology was analysed from radiographs using an active shape model and distinguished between fracture and control groups with an area under the receiver-operating curve (Az) of 0.81. Fourier analyses of different regions were compared and the best Az was 0.93. Neither shape nor structure measures were correlated with BMD, age or body-mass index (P>0.05). When these were combined with BMD, Az increased to 0.99. Further experiments and simulations showed that it might be possible to obtain both shape and texture measures from dual energy x-ray absorptiometry. Combined with BMD, this would make an improved assessment of fracture risk available to clinicians without requiring additional equipment.