Research On Mandibular Defect Model Reconstruction Based On Deep Learnin

The mandible is the only movable bone in the skull,which is responsible for the important functions of chewing food and speaking.When a large bone defect occurs,surgical treatment is required.Personalized 3D implant repair method has become a new development direction for mandibular defect repair.Efficiently designing and manufacturing implants that highly match the shape of the defect is of great significance for improving the success rate of defect repair surgery and ensuring the postoperative recovery of patients.Based on the deep learning method,this paper focuses on the automatic generation of a 3D model of the defect implant that is highly consistent with the patient’s mandibular defect and the optimization of the molding process parameters.The main research content of this paper is as follows:(1)Research on the automatic segmentation method of mandible CT images.An improved Unet network that combines attention mechanism and residual structure is proposed.The attention mechanism is used to enhance the network’s ability to extract key information,and the residual structure solves the problem of gradient explosion and gradient disappearance in deep networks.Experiments were carried out on the self-built mandible CT image dataset,and the MIo U of the improved Unet network reached 94.86%,and the segmentation accuracy of the mandible region was higher than that of the Unet and other mainstream image segmentation networks before improvement.(2)3D model reconstruction of mandibular defect based on CT image segmentation results.A GRD algorithm capable of generating random defects is proposed.Based on the complete mandible 3D model,a 3D mandible model with defects is generated,and a mandible 3D model dataset for defect repair is constructed.An improved 3D Unet network combining atrous convolution and residual structure is proposed to realize the automatic repair of the defective mandible and the generation of implant models.Based on the experimental comparison of the constructed dataset with the classic 3D Unet network and other networks,the Dice,Io U,PPV and Recall of the generated implants reached 0.8018,0.6731,0.7782 and 0.8330,respectively.Through the actual case of mandibular defect patients,the verification results show that the method proposed in this paper is feasible and effective.(3)Based on the generated defect implant models,the influence of key process parameters of polyetheretherketone(PEEK)implant fused deposition modeling(FDM)on the dimensional accuracy of implants was studied.Orthogonal experiments were designed,and the influence degree and significance of the four factors of printing speed,layer thickness,printing temperature and filling rate on the accuracy of implants were obtained through range analysis and variance analysis.The optimal combination of parameters was determined as printing speed 40mm/s,layer thickness 0.3mm,printing temperature 420℃and filling rate 80%,and the molding process parameters were optimized.