Research On Key Technologies Of Medical Image Guided Lumbar Surgery System

Minimally invasive spine surgery is an effective method for the treatment of lumbar spine disorders and is widely used because of its advantages of minimal tissue damage,short postoperative recovery period,and few complications.However,the small incision during the percutaneous puncture stage and the problem of non-visibility of anatomical structures such as subcutaneous vertebrae and tissues affect the safety and reliability of minimally invasive spine surgery.The imageguided surgical navigation system solves the problem of invisible subcutaneous anatomy by capturing and displaying the relative positions and postures of surgical instruments and the patient’s target surgical site in real time,which can assist the surgeon to complete minimally invasive spine surgery accurately and efficiently.However,due to the special anatomical structure of the lumbar spine and the intraoperative postural changes of patients,there are still great challenges in the actual clinical application of image-guided surgical navigation systems.Therefore,it is of great clinical significance to study the key technologies in minimally invasive lumbar spine surgical navigation systems.In this paper,we systematically investigate the lumbar image segmentation,lumbar image centroid localization,and lumbar image registration involved in medical image-guided minimally invasive lumbar surgery systems.First,for the problems of poor quality and low efficiency of lumbar spine CT image segmentation caused by complex and variable vertebral topology,irregular contour boundary,and noise,etc.,this paper proposes a vertebral CT image segmentation algorithm based on3 D V-network.The algorithm firstly improves the contrast of CT images by windowing technique,combines Dice loss and cross entropy loss to improve the stability of network training,and optimizes the output of the segmentation model by using morphological techniques.The experimental results show that the proposed method can achieve high quality segmentation of multi-scale vertebral images.To achieve accurate localization of the lumbar vertebral centroids,an automatic U-Net 3+ lumbar spine localization algorithm based on relative position constraints is proposed in this paper.The algorithm achieves image enhancement by means of image fusion and introduces the lumbar spine anatomical structure information into the U-Net 3+ network model through PCLoss to automatically locate the lumbar vertebral centroid and intervertebral disc centroid.The experimental results show that the proposed method can effectively learn the anatomical structure information of each centroid,and thus can locate the lumbar vertebral centroids and intervertebral disc centroids with high accuracy.In order to realize the accurate registration of lumbar spine images,this paper adopts the Trimmed ICP point cloud registration algorithm to realize the two-stage lumbar spine registration based on the characteristics of local rigidity and overall flexibility of lumbar spine.The method was used to obtain the initial registration matrix and the fine registration matrix.The experimental results show that the proposed method can efficiently achieve accurate registration of multimodal 3D lumbar spine images.