An Active Fracture Diagnosis System With Biocompatible Sensor Based on LSTM-MLP Quantitative Analysis

Traditional fracture diagnosis relies solely on medical images, which have limited diagnostic effectiveness. This article proposes a biocompatible sensor using fiber Bragg grating (FBG) to actively and precisely measure microstresses following fractures, advancing diagnostic approaches without the need for passive medical imaging. Additionally, a dual-scale LSTM-MLP hybrid temporal model that integrates a long short-term memory (LSTM) network and a multilayer perceptron (MLP) is proposed to train and evaluate strain from FBGs for a comprehensive analysis of bone injury conditions and guidance in the rehabilitation process. Compared to the convolutional neural networks (CNNs) that rely on medical images, our LSTM-MLP model requires fewer training instances and less computation time and can capture the entire dynamic process of fracture healing. The experimental results indicate that the biocompatible sensor has high sensitivity for measuring bone and plate strain with the values of 1.77 pm/N and 0.97 pm/N, respectively, and high linearity with the values of 0.998 and 0.997, respectively. Moreover, the LSTM-MLP model can assess fracture rehabilitation status and recurrence degree precisely by analyzing bone strain throughout the rehabilitation process with accuracy rates of 92.56% and 92.61%, respectively. Furthermore, the model can also provide effective predictions for subsequent stages of diagnosis and provide scientific real-time advice to patients.