Registration of Free-Breathing Abdominal 3D Contrast-Enhanced CT

Abstract

Abstract. CT perfusion imaging is used for the follow-up of abdominal tumors. A specificity of our work is that patients are breathing freely during image acquisition (5 minutes). We propose an automatic 3D image registration to compensate respiratory motion. The registration is computed in two main steps: global translation in the z-direction and 3D multiresolution blockmatching. Within this algorithm, the choice of similarity measure largely determines the algorithm robustness in presence of intensity shifts due to contrast diffusion. We exploit a modified entropy-based similarity measure to improve the quality of registration. We also propose two relevant criteria allowing to quantify the registration quality: one based on the gradients of image volumes of differences and one based on the smoothness of enhanced-intensity curves.

Registration of Free-Breathing 3D+t Abdominal Perfusion CT Images via Co-segmentation

“…We thus evaluated our registration by computing the Dice coefficient between the lesion in each frame and the lesion in the reference frame. Figures 3 shows this score, in comparison with the original sequence and the sequence registered by the block-matching method of [5]. Our method globally provides more precise and robust registration in the area of the tumour as the obtained Dice coefficients have both a higher mean and a lower variance.…”

“…Nonrigid iconic registration methods [2][3][4] are computationally demanding and rely on their similarity criterion. Standard choices such as mutual information are not effective in DCE-CT sequences [5]. In [6], Bhushan et al used the pharmacokinetic model fitting error as registration criterion for DCE-MR, thus coupling the two tasks of sequence stabilisation and parameter estimation.…”

“…Concerning free-breathing DCE-CT registration, previous work is limited -to the best of our knowledge -to [5] in which Romain et al proposed a registration by block-matching with a modified entropy-based similarity measure. This shall be considered as the baseline method.…”

“…In Section 3, we present results in terms of registration accuracy and parametric images and compare our approach to the baseline method [5]. Finally, Section 4 provides some discussion and concludes the paper.…”

“…3. Boxplots of lesions Dice coefficients from original sequences (red), sequences registered with entropy-based block matching [5] (blue) and the proposed method (green). kidney) would be stable along the frames.…”