Total knee arthroplasty radiographic evaluation via a Bayesian belief network. A pilot study

Catalin Cîrstoiu^{1, 2}, Bogdan Crețu^{1, 2,* ✉}, Bogdan Șerban^{1, 2}, Sergiu Iordache^{1, 2}, Adrian Cursaru^{1, 2}, Stefan Cristea^{1, 3}, Anca I. Eremia^{1, 4}, Silvia Nica^{1, 4}

1 - “Carol Davila” University of Medicine and Pharmacy, Bucharest, Romania
2 - Department of Orthopedics and Traumatology, University Emergency Hospital, Bucharest, Romania
3 - Department of Orthopedics and Traumatology, Pantelimon Emergency Hospital, Bucharest, Romania
4 - Department of Emergency, University Emergency Hospital, Bucharest, Romania

Correspondence: Bogdan Cretu

Received: July 10, 2020

Accepted: October 23, 2020

Abstract:

In an attempt to standardize the X-ray analysis and to create relevant correlations between it and the functional results, we present a pilot model of an artificial neural network based on a Bayesian belief network (BNN) designed for the automatic analysis of anteroposterior X-ray of the knee following total knee arthroplasty and more precisely the analysis of the tibial component. A prospective analysis was conducted in which 30 patients were analyzed and two groups resulted: the first was made up of 12 cases considered “normal” by the examiners (optimal positioning and cementation) and the second one, which was considered “pathological” (incorrect positioning or cementation) and included 16 cases. Based on 22 points established on the tibial component several geometric sizes (calculated parameters) were computed. Several parameters were compared and the results were up to 75% accuracy, 81.25% sensitivity, and 66.7% specificity. BBNs can be a solution for X-ray evaluation of total knee arthroplasty and can improve the understanding of its evolution. It needs confirmation in time, its accuracy rising with the number of patients introduced in the analysis algorithm.