THE EFFECT OF ANATOMICAL LANDMARK SELECTION ON REGISTRATION ACCURACY IN COMPUTER-ASSISTED ELBOW SURGERY

Abstract

Accurate implant alignment with the flexion-extension axis of the elbow is likely critical for optimal function and durability following elbow replacement arthroplasty. Implant alignment can be optimised by imaging the contralateral normal elbow prior to surgery and transferring this information to the diseased elbow in the operating room through registration. Successful registration is dependent on the presence of unique anatomical landmarks. Bone loss can create a challenge for registration as key anatomical landmarks are absent, limiting the number of sampling areas. This study investigated the effect of intraoperative sampling area on registration accuracy. We hypothesised that a low registration error can be achieved by acquiring surface data from areas unlikely compromised due to injury and readily available to the surgeon during typical surgical exposures.

CT images of twenty cadaveric distal humeri were acquired. Surface data was acquired from nineteen anatomical landmarks of the distal humerus using a hand-held laser scanner (FastSCANTM, Polhemus). Registration to the CT image was performed for thirty-nine landmark combinations. Only six combinations are discussed for succinctness.

Combining data from the anterior articular surface and humeral shaft, the lowest registration error was achieved in translation (0.8±0.3 mm) and rotation (0.3±0.2°). However, using data from the posterior shaft and proximal medial supracondylar column, a registration error of 1.1±0.2 mm and 0.4±0.2° was achieved.

Based on the results of this study, a low registration error can be achieved by acquiring data from two areas that are located proximal to the articular surface (the proximal medial supracondylar column and posterior humeral shaft), readily available surgically, and unlikely compromised due to distal humeral fractures, non-unions or bone loss due to severe erosive arthritis. Registration error was similar to the reported resolution of the laser scanner. Overall, this study demonstrates the promise for a successful registration of the contralateral normal elbow to physical surface data of the diseased or injured elbow using only a small portion of undamaged bone structure.