Revealing the evolution law of wear conditions and mechanisms is critical for assessing the health status of artificial joints and thereby extending their service life. Notably, high-precision wear particle recognition plays a pivotal role in the intelligent wear analysis of joint implants. However, insufficient and imbalanced particle samples often lead to inadequate recognition accuracy, which fails to meet the requirements of reliable wear particle analysis. Such limitations highlight the urgent need for a robust solution to address sample scarcity and imbalance. To this end, an intelligent wear particle recognition method based on a generative adversarial network is proposed to enhance the accuracy of artificial joint wear analysis. Specifically, a multi-scale coordinated attention-based generative adversarial network is designed to produce high-quality wear particle samples, which are used to supplement scarce sample categories and balance the particle distribution. Subsequently, a wear particle classifier is con-structed using a neural network, and the training strategy for this classification model is further optimized to ensure high-precision wear particle recognition. Ablation experiments demonstrate that the proposed method achieves an accuracy, precision, recall, and F1-score of 0.90 for wear particle recognition when using the generated samples. All metrics represent a significant improvement compared to the results obtained from the original sample dataset. Meanwhile, the recognition performance of the proposed method outperforms that of other state-of-the-art algorithms. Based on the recognition results of wear particles from joint implants, the particle size and type distributions are derived to evaluate the wear evolution of artificial joint. These findings hold great value for advancing artificial joint wear analysis and enabling reliable failure prediction.
Peng et al. (Sun,) studied this question.