Enhancing hip fracture risk prediction by statistical modeling and texture analysis on DXA images

Each year in the US more than 300,000 older adults suffer from hip fractures. While protective measures exist, identification of those at greatest risk by DXA scanning has proved inadequate. This study proposed a new technique to enhance hip fracture risk prediction by accounting for many contributing factors to the strength of the proximal femur. Twenty-two isolated cadaveric femurs were DXA scanned, 16 of which had been mechanically tested to failure. A function consisting of the calculated modes from the statistical shape and appearance modeling (to consider the shape and BMD distribution), homogeneity index (representing trabecular quality), BMD, age and sex of the donor was created in a training set and used to predict the fracture load in a test group. To classify patients as "high risk" or "low risk", fracture load thresholds were investigated. Hip fracture load estimation was significantly enhanced using the new technique in comparison to using t-score or BMD alone (average R² of 0.68, 0.32, and 0.50, respectively) (P < 0.05). Using a fracture cut-off of 3400 N correctly predicted risk in 94% of specimens, a substantial improvement over t-score classification (38%). Ultimately, by identifying patients at high risk more accurately, devastating hip fractures can be prevented through applying protective measures.

Enhancing hip fracture risk prediction by statistical modeling and texture analysis on DXA images Journal Articles