In International orthopaedics ; h5-index 43.0
PURPOSE : Acoustic analysis has recently been applied to cementless total hip arthroplasty (THA). The aim of this study was to develop a machine learning algorithm to predict post-operative subsidence with high accuracy.
METHODS : The acoustic parameters of the hammering sounds during a broaching procedure for 62 hips in 55 patients who underwent THAs with cementless taper-wedged stem were analysed. The patient's basic background such as age, sex, height, weight and body mass index, the femoral morphological parameters and the hammering sound characteristics of 24 features of normalised sound pressure (nSP) in 24 frequency ranges were applied to binary classification using a support vector machine using the following models with different features: model A, nSP only; model B, nSP + patients' basic background features; model C, nSP + patients' basic background features + femoral morphological parameters.
RESULTS : In 62 hips with 310 hammering sounds, 12 hips (19.4%) showed ≥ 3 mm of post-operative subsidence; hence, 60 hammering sounds were set as positive examples and 250 hammering sounds were set as negative examples. The AUC was very high in all models. The accuracy (AUC/sensitivity/specificity/positive predictive value/negative predictive value/accuracy rate) of each model was as follows: model A, 0.963/0.656/0.996/0.980/0.925/0.934; model B, 0.9866/0.675/1.000/1.000/0.928/0.937 and model C, 0.998/0.750/1.000/1.000/0.950/0.957.
CONCLUSION : In this study, we developed a high-accuracy machine learning algorithm for post-operative subsidence using acoustic parameters and additional pre-operative features. Our results represent a step toward the realisation of acoustic monitoring to avoid the complication in cementless THA.
Homma Yasuhiro, Zhuang Xu, Ohtsu Hiroshi, Ishii Seiya, Shirogane Yuichi, Hayashi Koju, Watari Taiji, Baba Tomonori, Ishijima Muneaki
2022-Nov-22
Acoustic analysis, Machine learning, Subsidence, Support vector machine, Total hip arthroplasty
