Diagnosis Of Lumbar Disc Herniation Disease Based On Convolutional Neural Network

Lumbar disc herniation(LDH)is a common spinal disease that usually occurs in the lumbar region.Traditional LDH diagnosis is often done manually,which is not only timeconsuming and labor-intensive,but also prone to misdiagnosis after working for a long time.LDH diagnosis based on deep learning has gradually developed due to its fast and accurate advantages,but it still faces some challenges.This thesis studies the diagnosis of LDH based on Convolutional Neural Networks(CNN),and uses spine imaging data to verify the diagnosis of LDH.The main work and innovations are as follows:(1)To solve the problem that the traditional diagnosis model of lumbar disc herniation has slow detection speed and can’t meet the time requirement of assisting doctors in the clinical diagnosis process,a deep learning model named SCENet is proposed.The model uses self-calibrating convolutions to explicitly expand the receptive field of each convolutional layer through internal communication to incorporate richer feature information,then the channel attention mechanism is used to learn the importance of each channel adaptively,and the channel is weighted according to its importance to improve the model performance.In the experiment on LDH data set,the prediction accuracy of SCENet model reaches 98.04%,fps reaches 8 and AUC reaches 0.96.Experiments show that the SCENet model performs well in terms of model detection speed and accuracy,and can detect lesions quickly and accurately.(2)To solve the problem that the traditional diagnosis model of lumbar disc herniation has low detection accuracy and can not meet the corresponding accuracy requirements when assisting doctors in clinical diagnosis,a detection model named MSKNet is proposed.This model uses a multi-scale module to solve the problem that the number of branches in the single-scale model is too small to fully extract feature information,then SK convolution is used to adaptively adjust the size of its receptive field,and Leaky Re LU activation function is used to replace the traditional Re LU activation function.In the experiment conducted on the LDH dataset,the prediction accuracy of the MSKNet model reached 98.31%,and the AUC value reached 0.97.Experiments show that the MSKNet model performs well in terms of detection accuracy,and can accurately detect and classify lesions.At the end of this thesis,a LDH aided diagnosis system is designed and developed by combining the proposed SCENet model with GUI interface,and good application results are obtained.