Image analysis for extracapsular hip fracture surgery

Abstract

During the implant insertion phase of extracapsular hip fracture surgery, a surgeon visually inspects digital radiographs to infer the best position for the implant. The inference is made by “eye-balling”. This clearly leaves room for trial and error which is not ideal for the patient. This thesis presents an image analysis approach to estimating the ideal positioning for the implant using a variant of the deformable templates model known as the Constrained Local Model (CLM). The Model is a synthesis of shape and local appearance models learned from a set of annotated landmarks and their corresponding local patches extracted from digital femur x-rays. The CLM in this work highlights both Principal Component Analysis (PCA) and Probabilistic PCA as regularisation components; the PPCA variant being a novel adaptation of the CLM framework that accounts for landmark annotation error which the PCA version does not account for. Our CLM implementation is used to articulate 2 clinical metrics namely: the Tip-Apex Distance and Parker’s Ratio (routinely used by clinicians to assess the positioning of the surgical implant during hip fracture surgery) within the image analysis framework. With our model, we were able to automatically localise signi cant landmarks on the femur, which were subsequently used to measure Parker’s Ratio directly from digital radiographs and determine an optimal placement for the surgical implant in 87% of the instances; thereby, achieving fully automatic measurement of Parker’s Ratio as opposed to manual measurements currently performed in the surgical theatre during hip fracture surgery.