基于自适应薄板样条全变分的肺CT/PET图像配准

CT/PET Lung Registration Using Adaptive Thin Plate Spline-Based Total Variation Regularization

全变分正则项虽然能够在具有滑移运动的肺等胸腹部器官图像配准时校正边界不连续位移场,但仍然无法保留图像的局部特征,损失配准精度。针对肺图像CT单模配准和CT/PET双模配准,通过像素点空间位置权重将薄板样条能量算子与全变分算子进行空间加权建立自适应薄板样条全变分正则项。然后,将正则项与CRMI相似性测度以及L-BFGS优化方法结合建立非刚性配准算法。通过DIR-Lab 4D-CT公共数据集和CT/PET临床数据集上的实验结果表明,提出的方法能够在保证边界不连续运动的同时保证图像内部的平滑性,具有更高的配准精度。

The discontinuities displacement field at the boundaries is preserved through Total Variation(TV)regularization when registering organ images with sliding motion.But TV assumes a global regularization which will lead poor correspondences at local areas of images.This paper creates adaptive Thin Plate Spline-based Total Variation(TPS-TV)regu-larization combining Thin Plate Spline(TPS)and TV operator according to the distance of pixels to boundaries.Then this paper chooses Correlation Ratio-based Mutual Information(CRMI)similarity measure function and Limited-memory Broyden Fletcher Goldfarb Shanno(L-BFGS)optimization to establish a non-rigid registration frame.Compared with TV and TPS regularization on the public DIR-Lab dataset and clinical CT/PET dataset,the proposed method has been demonstrated a more dependable displacement field and higher registration accuracy.