Injection Technology Evaluation And Injection Area Localization For Botulinum Toxin Type A Injection Into Masseter Based On CT Images

Masseter hypertrophy refers to the asymptomatic enlargement of unilateral or bilateral masseter,which usually results in a square jaw contour,which is generally considered unaesthetic in Asian.Today,Botulinum Toxin Type A injection therapy for masseter muscles has been widely used in Asia due to its low invasiveness and good therapeutic effect on masseter hypertrophy.However,the mainstream BoNT-A masseter muscle injection technology is basically summarized from clinical practice,and there has not been any research on credible quantitative assessment of masseter injection technology based on anatomical structure information.The CT image data of the head contains anatomical information of masseter muscles and other related anatomical structures,and it is easier to obtain than anatomical samples.Therefore,it is very meaningful idea to use CT images to evaluate BoNT-A injection technology.In addition,according to medical research related to facial reconstruction,a person’s facial contour implicitly encodes information such as the position and shape of the masseter,so locating the injection area of the masseter based on the facial contour features of the person is a direction worth exploring.This article focuses on the following three aspects of BoNT-A masseter injection technique evaluation and injection area localization:Firstly,a framework for mandible and masseter segmentation based on convolution neural network is proposed.The network is a novel multitasking framework,in which Ro I positioning and intra-area segmentation share the same backbone encoder network.Combined with the region proposal given by the encoder,multi-level Ro I region features are cropped from the encoder to form a decoder with high GPU memory efficiency to keep the segmentation details,thus expanding the applicable size and effective receptive field.In order to train the model more efficiently,a loss function based on Dice is designed for the global to local multitask learning process.Finally,the segmentation experiment of masseter muscle and mandible is carried out with53 cases of head CT image data,which is obviously better than the traditional advanced frame V-Net in terms of accuracy and efficiency.Secondly,the three-dimensional visualization and measurement of masseter muscle and mandible are realized.The system includes a complete3 D visualization rendering pipeline based on CT images,which can visualize injection-related anatomical structures such as masseter,mandibles,etc.,and provides injection path simulation and masseter thickness measurement based on a ray-casting method which are used for subsequent research.In addition,the system also enables interactive annotation of anatomical landmarks based on 3D shapes in order to produce annotation data needed for subsequent research.Thirdly,the anatomical basis of BoNT-A masseter injection was analyzed based on CT images,and the mainstream BoNT-A masseter injection technique was quantitatively evaluated.Based on the three-dimensional mesh representation of masseter muscle,mandible and facial skin,we propose a set of quantifiable masseter muscle based anatomical benchmarks based on rules summarized from the clinical medical literature,and use computer graphics techniques to quantitatively simulate Several sets of mainstream masseter muscle injection techniques were evaluated and evaluated based on anatomical benchmarks.Finally,it was concluded that the masseter muscle injection technique based on the cheilion-earlobe reference line is currently the best clinically.Finally,the deep learning method is used to locate the injection area.Based on the Point RCNN network,we treat the localizationas a 3D object detection task,and using the injection area simulated by the masseter muscle injection technique based on the angle of cheilion-earlobe as the ground truth,we trained on 50 samples to detect the injection area point cloud from the human face 3D point cloud and the the network was evaluated.Because the connection between the three-dimensional features of the face and the injection area is quite obscure,it is a rather difficult task compared to the mainstream 3D object detection problem,but we still get a good performance.