Research And Development Of LDH Automatic Diagnosis System Based On Deep Learning

At present,the diagnosis of lumbar disc herniation relies on the subjective judgment of the doctor,and the accuracy of diagnosis results will be limited and affected by the doctor’s experience and knowledge level.Therefore,it is of great significance to realize the automatic diagnosis of lumbar disc herniation.Based on reading a large number of related literatures on spine disease diagnosis technology,this thesis researched and developed an automatic diagnosis system for lumbar disc herniation based on deep learning.According to the degree of degeneration of the intervertebral disc,the position and size of the protrusion,the Pfirrmann classification diagnosis model and the MSU classification diagnosis model are realized,and the automatic diagnosis system for lumbar disc herniation is constructed to provide diagnosis basis and advice for patients and doctors.Intervertebral disc degeneration is the main cause of lumbar disc herniation.Pfirrmann classification is used to evaluate the degree of intervertebral disc degeneration.First,in order to extract the region of interest,that is,the intervertebral disc image,the ResNet algorithm is used to locate the intervertebral disc and the center point of the vertebral body in the sagittal image.In order to avoid positioning errors,the positioning constraints of intervertebral disc and vertebral body center are positioned based on the prior knowledge of spine anatomy,and an iterative correction algorithm is proposed to obtain the correct center point coordinates.Based on this,the cropped disc image is sent to the Pfirrmann classification classifier,and obtain the diagnosis results.To determine the size and position of the disc herniation,namely MSU classification diagnosis,images of the disc and part of the spinal canal were cut to make the herniation diagnosis and obtain the herniated disc based on the center point coordinates.In order to obtain the auxiliary line of MSU classification,the key points of the axial graph were located based on ResNet algorithm.In order to improve the positioning accuracy and correct the results of positioning errors,the constraints of key points positioning were proposed based on the spatial geometry of the spine,and the key point iterative correction algorithm is proposed to correct the wrong coordinates.Based on U-Net network,the segmentation algorithm of intervertebral disc and herniation was designed to obtain the herniation position,determine the position relationship between herniation and auxiliary line,and obtain the MSU classification and diagnosis results.Finally,an automatic diagnosis system for lumbar disc herniation was designed and developed.The system adopts the SpringBoot technology framework,Thymeleaf template engine,combined with Flask to deploy the deep learning model,and develops the application architecture of the B/S model.The use of multi-threaded methods to process the doctor’s immediate diagnosis and background data queue reading and diagnosis,combined with actual cases to verify the efficiency and feasibility of the system.