Research On Vertebral Block Segmentation Algorithm In MR Images

In vertebral MR images,the pathological signal of the patient’s body is usually hidden in the displacement of the lumbar position,the change of the lumbar size,or the change information of the intervertebral disc.Developing an algorithm of vertebrae segmentation and realizing the automatic segmentation system of vertebrae under multiple modalities in MR images will be helpful for doctors’ diagnosis and treatment.The algorithm is divided into three modules,namely,the preprocessing of vertebral image,the segmentation of vertebral blocks based on deep learning and the repair of prediction results.In preprocessing,the optimal layer is selected from multi-level sagittal images based on the characteristics of regional integrity of spinal cord and edge clarity of vertebral block;image enhancement is realized based on gray value stretching and anti-sharpening algorithm;vertebral segment position is judged according to the concave and convex of the right side line of vertebra and vertebral width information,and region of spine is extracted for deep learning and result optimization.The vertebral region extracted from multiple modal data is mixed together and trained by V-Net network model,and the loss function is soft dice loss function.The training effect of region image is better than that of whole image.According to network model,the vertebral image obtains initial segmentation results.The optimization process is divided into two parts: detection and segregation of adhesive vertebral mass and detection and repair of missing vertebral mass.The detection of adhesive vertebral masses utilizes the height and area information of multiple vertebral masses after adhesions.For hollow adhesions,the holes are separated according to the vertical line;for unilateral adhesions,the holes are separated from the concave points detected by the convex hull algorithm.Detection of missing lumbar masses is mainly based on the size and width of the lumbar masses.Finally,third part is talking about the repairment of missing lumbar block.For image that has obvious intervertebral discs,the intervertebral disc is cut out from thespine region.For the left images,they are repaired by region growth algorithm.Shape fitting method is used to repair the missing vertebral block for the vertebral mass with poor repair effect.This algorithm obtains very accurate vertebrae segmentation results in image data provided by Wuhan Lian Ying Science and Technology Co.The results show that this algorithm can accurately segment the vertebral blocks from the multi-modal vertebral images.