Research On Detection Algorithm Of Lumbar Nerve Root Nuclear Magnetic Resonance Image

Lumbar disc herniation has become one of the more common diseases.There are two main clinical conditions that can cause this disease.One reason is that after various degrees of degenerative changes in various parts of the lumbar disc,the fiber ring of the lumbar disc ruptures under the action of external forces,And nucleus pulposus protrudes from the rupture to the back or inside the spinal canal,causing adjacent spinal nerve roots to be stimulated or compressed,resulting in a series of clinical symptoms such as lumbar pain,numbness and pain in one or both lower limbs.Another reason is that the enlarged varicose veins in the lumbar vertebral canal compress nerve roots,causing pain in the lower back and causing a series of clinical symptoms such as numbness in the lower extremities.The lumbar and leg pain caused by the abnormal varicose veins in the spinal canal is difficult to analyze and confirm through the original MRI image.Therefore,this paper mainly studies the problem of lumber nerve root detection in the MRI image.There are several difficulties in the detection of lumber nerve root MRI images,such as polymorphic,small target and unbalanced datapolymorphic means that the nerve roots of the same part show different states at different times and we propose a DSSD(deformable convolution Single Shot Detector)target detection algorithm to solve the problem of polymorphic.At the same time,for the problem that the DSSD convolutional neural network model does not detect small targets accurately,we propose an improved model of DSSD based on the feature pyramid structure.The feature pyramid network can fuse deeper convolutional feature maps with more abstract and richer semantic information and shallower convolutional feature maps with higher resolution and more details.Finally,in order to deal with the problem of data imbalance,we dynamically adjust the cross-entropy loss according to the confidence level,and improve the standard cross-entropy loss function.We first test the detection algorithm proposed in this paper on the standard data sets VOC2007 and VOC2012,and compare it with the SSD target detection algorithm.The experimental results show that our proposed algorithm has better performance.Then we tested it on the clinical MRI image dataset and compared it with other models.The results showed the superiority of the lumber neural root multimodal detection algorithm.