A CNN-LSTM deep learning model achieved 92.3% accuracy in predicting upper crossed syndrome, outperforming the best single-LSTM model by 4.5 percentage points.
Does a deep learning-driven CNN-LSTM model improve risk prediction accuracy for upper crossed syndrome compared to single-LSTM models?
A novel CNN-LSTM deep learning model improves the accuracy of risk prediction and early warning for upper crossed syndrome compared to traditional single-LSTM models.
Tasa de eventos absoluta: 92.3% vs 87.8%
As a result of unhealthy lifestyles that include spending increased time sitting at office and staring at their computer screens, cases of upper crossed syndrome continue to increase year after year with serious consequences on the muscle performance, physical appearance, and quality of life of individuals. The importance of early risk prediction and early warning is of much value in disease prevention and control. The existing risk evaluation of the upper crossed syndrome is largely based on the conventional scales or manual observation of physical symptoms that are characterized by high subjectivity, low efficiency, and delayed early warning and therefore, it is hard to obtain accurate and early prevention and control. The structure of this paper is as follows: First, it will conduct a review of the factors that influence the upper crossed syndrome, and some of the current assessment methods to establish the essential requirements to construct a model. Second, it builds a risk prediction and early warning model on the basis of deep learning, enhances the quality of data with a multimodal data (postural parameters, lifestyle habits, electromyographic signals, etc.) preprocessing module, and fuses multi-dimensional features with a better convolutional neural network-long short-term memory network (CNN-LSTM) to establish an adaptive threshold early warning mechanism. Lastly, it performs comparative experiments on clinical datasets and simulated datasets to check the performance of the model. Experiments have demonstrated that the proposed fusion model is the most effective in terms of all the evaluation metrics with the highest accuracy of 92.3% which is 4.5 percentage points high in comparison to the best single-LSTM model. The precision and recall are 91.8 and 93.1 respectively as well balanced between the problems of "overlooking high-risk cases" and "false identification of healthy cases" respectively, and have reliable technical support in early intervention in the upper crossed syndrome.
Du et al. (Thu,) conducted a other in Upper crossed syndrome. CNN-LSTM deep learning risk prediction model vs. Single-LSTM model was evaluated on Model accuracy. A CNN-LSTM deep learning model achieved 92.3% accuracy in predicting upper crossed syndrome, outperforming the best single-LSTM model by 4.5 percentage points.