Research On 3D Reconstruction And Spatial Registration Algorithm Of Brain And Scalp

Transcranial Magnetic Stimulation(TMS)is a physical treatment for related disorders of the brain.This approach has been widely used in clinical treatment in recent years due to its remarkable therapeutic effect,as well as non-invasive and selective stimulation.During TMS treatment,the spatial localization of stimulation target on the scalp is necessary to ensure safe and effective treatment.However,the existing localization methods rely on the head-mounted auxiliary locator and manual registration,which to some extent,increase the operational complexity,bring a worse treating experience and limit the flexibility of treatment.In order to solve these shortcomings,focusing on the 3D reconstruction of brain and scalp,as well as spatial registration algorithm,the thesis designs a novel and visual-based spatial localization system for the stimulation target,which can be applied in an unconstrained,markerless circumstance.Referring to the process of TMS treatment and the design objectives,this thesis proposes a scheme of the spatial localization system.According to the in-depth investigation on the characteristics and limitations of the traditional localization approaches,requirement analyses of the system including functional requirements of reconstruction and registration,as well as performance requirements of localization accuracy and real-time performance,are presented.On this basic,an overall solution of the mentioned system is constructed including processes,mathematical model,algorithm platform and hardware calibration.In order to realize the real-time spatial localization of the stimulation target,a spatial registration method based on 2D and 3D facial landmarks is proposed.For the construction of 3D reference for the registration,this thesis detects the face region from a single face image with combination of Haar-like features and Adaboost algorithm,identifies the facial landmarks in the region through the constrained local model and reconstructs 3D face efficiently by fitting a morphable model.Moreover,the 3D brain and scalp from the MRI are reconstructed by the marching cubes algorithm,where the grid simplification based on quadratic error measure and laplacian mesh smoothing algorithm are adopted to settle the problem of numerous point clouds as well as the rough surface.By position registration of the 3D scalp and brain,and adopting the fast nearest neighbor algorithm,the pre-selected stimulation target on the scalp that is closest to the intracranial point on the brain can be determined.After the above process,the thesis proposes a 3D facial landmarks localization method by combining the curvature with spin image templates of the landmarks,and realizes the spatial registration of the face and scalp by the alignment of their corresponding landmarks.According the chain registration,based on the RGB and optimized depth images in the video stream,real-time spatial localizations of the stimulation target with respect to the camera is realized by estimating the head pose and position.With the spatial localization system built in this thesis,experiments are conducted by comparing the results with the measurements from the real-time motion tracking system.The effective accuracy and the real-time performance of the system are verified,and PC software relying on the Qt Creator is designed for visualization of the registration and localization processes.