Construction of an Artificial Neural Network Model for Predicting Ankle Ligament Injury Based on the Results of Ultrasonic Shear Wave Technology and Magnetic Resonance Imaging

Abstract

We aimed to study the factors influencing the extent of anterior talofibular ligament (ATFL) and calcaneal ligament (CFL) injuries in acute ankle fractures based on construction of an Artificial Neural Network (ANN) Model. Differences in various baseline data, including personal data, anthropometric data, disease history, and life history, were compared among patients with ATFL injury, ATFL+CFL injury, and ATFL fracture. Factors influencing the ATFL injury/ATFL+CFL injury/ATFL fracture were analyzed using logistic regression, and an artificial neural network (ANN) model for predicting ATFL fracture was constructed using the tensor flow framework. Advanced age (OR= 36.33, 95%CI (15.72, 60.62)), male (OR = 21.21, 95%CI (5, 39.92)), high BMI (OR = 0.03, 95%CI (−0.31 0.37)), exercise duration (OR = 0.48, 95%CI (−14.66, 18.3)), and history of diabetes (OR = 16.98, 95%CI (−76.44, 480.78)) may all be influential factors in the ATFL and CFL injury/ATFL rupture. We constructed three neural layers, the first containing 11 ganglia, the second containing 7 ganglia, and the third containing 5 ganglia, and after 10 iterations the ANN model LOSS values were reduced to the lowest and scatter plots were made of the true and predicted values with some linear trend and better prediction. Advanced age, male, high BMI, length of exercise, and history of diabetes may be influential factors in the ATFL and CFL injuries/ATFL rupture. Applying the tensor flow framework, the ANN algorithm was constructed to predict the occurrence of ATFL fracture with good results.