| ?Objective?The purpose of this study is to apply a fully automatic computer-aided diagnosis(CAD)system based on artificial intelligence,extract the intervertebral disc in the lumbar spine MRI image,diagnose the lumbar disc herniation,and analyze the Pfirrmann degeneration grade and MSU prominence degree classification of lumbar intervertebral disc.Calculate the positive rate of disease diagnosis,the accuracy,sensitivity,specificity and intersection-over-union of typing results,and the final result of human-machine war.To explore the advanced nature of the designed procedure and the necessity of CAD system,and diagnose the spine for artificial intelligence.Provides a basis for the diagnosis of other diseases on spine by artificial intelligence,and provides new ways to improve clinical work efficiency.?Method?The MRI imaging data of 217 patients with lumbar disc herniation were divided into training and experimental groups.The artificial intelligence software designed and trained was used to perform the vertebrae positioning using the Faster Region Convolution Neural Network,and then the intervertebral disc was performed through the vertebrae.Positioning and selection.The diagnosis of disc herniation was performed using a binary neural network classifier for the selected area.A computer algorithm for identifying three image features using histogram of pixel intensity(HPI),local binary pattern(LBP)and pyramidal histogram of oriented gradients(PHOG),and Pfirrmann segmentation of the intervertebral disc based on the multi-layer perceptron classifier type.The keypoint detection neural network is used to locate the marker points in the MSU definition.The U-Net network structure is used for calculation,and the MSU is classified.Finally,accuracy,sensitivity,specificity and Intersection-over-Union(IoU)are used as evaluation criteria,and man-machine battles are carried out to obtain experimental data and perform statistical analysis to determine the necessity of artificial intelligence assistance.?Result?We used half of the data from the MRI imaging data obtained by the Radiology Center of the Tianjin Medical University of General Hospital to verify the performance of our artificial intelligence in the diagnosis of lumbar disc.In the positioning of the intervertebral disc,the accuracy rate can reach 95%;in the diagnosis and detection of lumbar disc herniation,the accuracy rate is 95.83%,the specificity is 97.52%,the sensitivity is 94.12%;the Pfirrmann classification,the accuracy rate is 82%.In terms of MSU classification,the average error of key point detection is 3 pixels,and the average error of the segmentation process is 90%.In the human-machine war,the artificial intelligence in the diagnosis of lumbar disc herniation,MSU classification and overall time to make a diagnosis show a clear advantage.?Conclusion?Our group has proposed a fully automated CAD system that can be used as a good aid in the workflow of radiologists and clinicians.The MSU classification of lumbar disc herniation can help doctors to make more appropriate treatments for different types of patients.Conservative treatment is often used in patients with 1-A,1-B and 2-A types.Surgical treatment is often used in the 2-B and 2-AB types,and multiple treatments are needed for the treatment options of 1-AB and C.Pfirrmann classification of intervertebral disc degeneration,in theory,the higher the Pfirrmann level,the more difficult the surgery,providing a warning for clinicians,and the use of this classification in preoperative assessment can help patients with degenerative scoliosis Surgical decision making.For the purpose of clinical application,the function of CAD has a good performance.After extensive validation,the team's CAD system is expected to be applied in the clinic,and will solve the problem of accurate diagnosis and reasonable treatment of LDH in community hospitals or remote hospitals that lack professional doctors and low medical standards. |
PDF Full Text Request