Contribution to the clinical validation of a generic method for the classification of osteoarthritic and non-pathological knee function

• Main Author: Whatling, Gemma Marie
• Published: Cardiff University 2009
• Subjects: 612
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.584565

The Cardiff Dempster-Shafer (DS) classifier is a generic automated technique for analysing motion analysis (MA) data. It can accurately discriminate between level gait characteristics of non-pathological (NP) and osteoarthritic (OA) knee function. It can also quantify and visualise the functional outcome of a total knee replacement (TKR). A number of studies were undertaken to explore and enhance this method. The training set for the classifier was increased by 48% by collecting additional knee function data for level gait. Knee function for nine new patients was classified pre and post-TKR surgery. At 12 months post-TKR, two patients exhibited non-dominant NP knee function. The remaining patients did not recover NP gait. This finding is similar to previous classifications of level gait. To improve the distinction between varying degrees of knee function, stair gait was introduced into the trial. A staircase was designed and validated. Adduction and flexion moments acting about the knee joint and medial component of the ground reaction force were found to be important in the classification of OA and NP knee function from stair gait. Using a combination of these variables the DS classifier was able to characterise OA and NP function for 15 subjects correctly with 100% accuracy, determined using a leave-one-out method of cross validation. The variables were tested to assess the outcome of TKR surgery. The patient assessed recovered NP stair gait post surgery. An image based study was undertaken to investigate the quality of the MA data used in the DS classifier. A step up/down activity for 5 NP and 5 TKR subjects was recorded using non-simultaneous MA and dynamic fluoroscopy. Accurate knee kinematics were computed from the fluoroscopy images using KneeTrack image registration software. MA measured significantly larger knee joint translations and non-sagittal plane rotations. The largest errors in MA derived kinematics were 9.53 for adduction-abduction range of motion (ROM) measured from the NP cohort and 2.63cm compression-distraction ROM of the tibio-femoral joint, measured from the TKR cohort. The generic nature of the DS classifier was tested by its application to distinguish hip function following a lateral (LA) and posterior (PA) approach to total hip arthroplasty. The use of different variables was investigated with the classifier. The best classifier was able to distinguish between NP and LA function with 96.7% accuracy, LA and NP with 86.2% accuracy and between LA and PA with 81.5% accuracy. The PA approach was found to lead to more characteristic NP hip function than LA. These studies show that variables from stair gait should be included in addition to level gait in the classifier. Due to errors when measuring non-sagittal plane rotations using MA, these should be interpreted with caution. The generic nature of the classifier has been proven by its application to another joint, thus answering another orthopaedic question.