| Dental orthodontic treatment is mainly used to improve the alignment of patients’ teeth to improve their bite force.Orthodontic instruments such as arch wires are usually used,which will bring painful experience to patients.The evaluation of pain intensity is generally based on the patient’s score on visual analog scale(VAS),which is a subjective evaluation method,while seeking an objective pain evaluation program can be used as an auxiliary and supplement for pain evaluation.In addition,in the process of orthodontic treatment,common pain relief programs include drug therapy,cognitive behavioral therapy,music therapy,etc.,so it is of certain positive significance to analyze the mechanisms of different pain relief programs to improve the pain state.In this paper,EEG data of 36 patients undergoing orthodontic treatment were collected.EEG data,as a physiological signal,can reflect a person’s physiological health status and cognitive level.(1)Aiming at the problem of objectively assessing the pain degree of patients,this paper proposes to classify the pain degree of patients’ pain EEG data based on deep learning after orthodontic treatment.This paper proposed DFB-ARCNN model for the first time,and innovatively proposed the feature fusion scheme of "CBAM+ point-bypoint convolution".The model extracts features from EEG in different frequency bands,then assigns adaptive weights to features in different frequency bands by CBAM attention mechanism,and then integrates features in different frequency bands by pointby-point convolution.Finally,BI-LSTM is used to further extract time-dependent features of fused features.Compared with SOTA method in the field of pain EEG classification,the improved model DFB-ARCNN in this paper achieved the highest accuracy of 99.11% and 76.59% under subject dependence and subject independence,respectively.The DFB-ARCNN model with independent subjects can be used as an objective tool to evaluate the degree of orthodontic pain.(2)Aiming at the mechanism of pain relief of different treatment schemes,this paper classifies the EEG data of different pain relief schemes based on deep learning,and visualizes the model features to reveal the differences and significance of different pain relief schemes.The CNN-LSTM model and DFB-ARCNN model were constructed to classify the EEG data of different pain relief programs.Finally,the EEG data of cognitive behavioral therapy group(CBT),brainwave music therapy group(BM),and blank control group(CO)without pain treatment program were classified with 96.35%and 98.20% accuracy respectively.By visualizing the hidden layer features of the two models,the similarities of the extracted features of the two models were compared and analyzed,and combined with the related studies of EEG signals,the differences of different treatment programs were explained,that is,the PSD value of low-frequency EEG signals decreased in the CBT group and the BM group,which revealed that this feature may be related to pain perception.Finally,the effectiveness of the improved DFB-ARCNN model is explained by visually comparing the feature differences between CNN-LSTM model and DFB-ARCNN model.The EEG classification scheme adopted in this paper can objectively reflect the pain degree of patients with high accuracy.Deep learning was used to study the differences in pain relief between different pain relief programs,and visual display and computational analysis were performed to explain the differences in different pain relief programs. |
