Novel image processing techniques of standard DXA scans to better predict hip fracture risk

Osteoporosis is a generalized skeletal disorder common in older adults, in which a reduction in Bone Mineral Density (BMD) decreases bone’s strength and can result in an increased risk of fragility fractures. One of the sites commonly affected by osteoporosis is the proximal femur (“hip”), the fracture of which greatly decreases mobility and function, as well as being responsible for high health care costs for society (estimated at $619M annually)1,2.

We have developed novel image processing techniques to extract much more information from each clinical DXA scan than simply a BMD measure. This approach uses statistical modeling techniques to characterize the shape and distribution of mineral through the proximal femur, as well as the trabecular quality through texture analysis. A template model is established from a training set of images with patient fracture history data, after which any new scan may be mathematically represented by its differences from the template shape through scaling factors applied to the main modes of variation. The large complement of DXA scans in longitudinal databases presents an untapped opportunity for further investigation of our fracture-predicting algorithms and characterization of how different factors (such as medication use) contribute to fracture risk.

Faculty Supervisor:

Rick Adachi;Cheryl Quenneville

Student:

Fatemeh Jazinizadeh

Partner:

Amgen

Discipline:

Other

Sector:

Other

University:

McMaster University