摘要
Demons算法的一个局限是它无法处理大形变,且不能产生微分同胚的变换以满足计算解剖学的形态分析需要。利用李群中的指数映射,把原来Demons形变场相加的更新方式改进为若干次形变场间的复合,同时又保证了较高的运算效率。实验表明,新算法能配准大形变问题,且真实颅脑CT的实验结果与Demons算法的结果相近,但产生的形变场光滑可逆,并有相对更小的形变能量。
One of the limitations of the Demons algorithm is that it can neither handle large deformation nor generate diffeomorphic spatial transformations which are required for the shape analysis in the framework of Computational Anatomy. We proposed to replace the addition of free-form deformation in the Demons ＇ update step by a few compositions of deformation fields using group exponential. The algorithm proved to be computationally efficient. Our experiments showed that this new algorithm was capable of recovering large deformation and is diffeomorphic with similar results of the Demons algorithm. The generated deformation were smooth and invertible in terms of Jaeobian with much lower energy of the deformation fields.
出处
《北京生物医学工程》
2010年第1期49-54,共6页
Beijing Biomedical Engineering
基金
国家自然科学基金(60771007)资助
关键词
非刚性配准
微分同胚
指数映射
优化
non-rigid image registration
diffeomorphism
group exponential
optimization
