Nonrigid Point Set Matching Algorithms And Applications To Medical Images

With the rapid development of medical imaging devices and computer technologies,computer-aided simulation in medical image domain plays an increasingly important role.Non-rigid point set matching is one of the important computer-aided simulation technologies.Non-rigid point set matching can perform not only on 2D images but also on the reconstructed 3D volumes.Meanwhile,non-rigid point set matching can also be used to assist in making surgical planning,evaluating operation results,discovering and studying diseased tissues.Based on the study and analysis of current technologies of the cranio-maxillofacial opera-tion evaluation and heart DTI fiber atlas,this thesis focuses on probability-based non-rigid point set matching,and proposes several non-rigid point set matching algorithms suitable for medical images.Our main works and innovations are as follows.?1?Proposed a reference-omitted affine point set matching algorithm.This method effec-tively reduces the influences of different reference sets,especially when a large difference in scale exists between the two sets.With the alternate reference set and soft correspondence,the optimization avoids falling into a local minimum prematurely,and gets better matching results.?2?Proposed a robust coherent point drift approach based on rotation invariant shape con-text.With the rotation invariant shape context,this method enlarges the application of CPD algorithm to the situation of large rotations.With the adaptive prior and outliers ratio,this method can actively adapt to the changing point sets.The correspondences and non-rigid trans-formations can be decided based on the distances and shape features of point sets.?3?Built the framework of sparse non-rigid point set matching,and proposed two special algorithms,SNR-TPS and SNR-CPD.By introducing sparse matching error residues of?_p?p?[0,1]?norm,the proposed framework solves the problem that matching error residues of?₂norm is highly sensitive to outliers.All algorithms based on this framework can achieve the best local optimization with the deterministic annealing and alternating direction method of multipliers.All of the above methods can be used to evaluate cranio-maxillofacial operation results.These methods solve the different problems in point set matching.They can also be used in a combined manner,from coarse to fine,which makes it possible to quantify surgical operation.?4?Proposed heart fiber based joint clustering and matching algorithm to construct the atlas of heart DTI.Due to the constraint of fiber information,this method ensures fiber consis-tency and shape invariance when clustering heart fibers.With joint clustering and matching,this method reduces the quantity of data to handle and achieves the one-to-one correspondence between different sets.The construction of heart DTI atlas is the basis for studying heart fiber diseases and fiber feature of the human heart.