Spinal disorder diagnosis based on deep learning integrating biomechanical data

In the field of spinal disease prediction, various approaches, including regression analysis algorithms, traditional machine learning and some deep learning techniques, have been employed. While these methods have contributed to spinal disease diagnosis to some extent, they exhibit limitations such as a strong dependency on feature selection, high data requirements, and restricted scalability and generalization capabilities. To address these challenges, we propose a novel deep learning-based framework that integrates multiple enhancement algorithms into the Long Short-Term Memory (LSTM) architecture. By preprocessing the data and enabling the model to autonomously learn complex patterns and abstract features, the proposed approach effectively evaluates spinal disease data. Six biomechanical parameters were used as the input of the model, and Normal, Hernia and Spondylolisthesis were used as the output to judge the category. Our model demonstrates superior performance compared to other models, reaching an accuracy of 85.8%, 85% weighted precision, and 78.7% recall, providing an innovative deep learning solution for spinal disease diagnosis.