Research on Injury Identification Based on Analytical Neural Network

This paper presents a study on the identification of intraoperative injuries in equine colic surgery using an analytical neural network. Twenty-six features were employed to predict whether horses experienced an intraoperative injury. The results indicated an average MSE of 0.261 across multiple experiments. Furthermore, an analysis of the neural network revealed the importance of specific input features for predicting intraoperative injuries, highlighting the significance of the horses' final outcomes (including lived, died, and was euthanized). Based on the analytical results, a segmented neural network model was constructed, which initially classified the dataset using the identified crucial features and then predicted using a backpropagation (BP) neural network. The average MSE for the segmented network across three categories was found to be 0.1778, 0.1579, and 0.3727, with an overall average MSE of 0.2361. A comparison between the basic BP neural network and the segmented neural network demonstrated that the latter exhibited higher total prediction accuracy, particularly in the lived and died categories. Although the segmented network's accuracy in predicting the “was euthanized” category was inferior to the basic BP neural network's MSE across all categories, it displayed fewer outliers and greater prediction stability.