Using Machine Learning to Detect and Predict Loosening NexGen Total Knee Replacement

The goal of this trial is to investigate whether Machine Learning (ML) can be used to detect small degrees of loosening, lucent zones, or any other changes on radiographs that might predict early failure following NexGen total knee replacement.

Researchers will identify plain AP and lateral plain film radiographs from two groups of patients. Those who has NexGen total knee replacements (TKRs) that went on to failure, and those who has well performing TKRs. Radiographs from these two groups will be labelled as 'failure' and 'well performing' and will be processed through a machine learning algorithm.

The algorithm will be successful if it is able to detect a NexGen TKR that went on to failure or went on to perform well. This will be determined by using a test set.

The population will be adults who had the recalled a NexGen Total Knee Replacement with a standard tibial tray. It will include adults only, who has the TKR at University Hospitals Southampton between 2003 and 2022.

Failure will be defined as revision of tibial or femoral components which is likely due to aspectic loosening. It will exclude washouts, exchange of poly, peri-prosthetic fractures, microbiologically confirmed infection.

Well performing TKRs will be defined as patients who have had their TKR in situ for 10 years and have reported no significant symptoms.