Medical & Biological Engineering & Computing
https://doi.org/10.1007/s11517-018-1808-1
ORIGINAL ARTICLE
An atlas-based multimodal registration method for 2D images
with discrepancy structures
Wenchao Lv
1
· Houjin Chen
1
· Yahui Peng
1
· Yanfeng Li
1
· Jupeng Li
1
Received: 18 April 2017 / Accepted: 16 February 2018
© International Federation for Medical and Biological Engineering 2018
Abstract
An atlas-based multimodal registration method for 2-dimension images with discrepancy structures was proposed in this
paper. Atlas was utilized for complementing the discrepancy structure information in multimodal medical images. The
scheme includes three steps: floating image to atlas registration, atlas to reference image registration, and field-based
deformation. To evaluate the performance, a frame model, a brain model, and clinical images were employed in registration
experiments. We measured the registration performance by the squared sum of intensity differences. Results indicate that
this method is robust and performs better than the direct registration for multimodal images with discrepancy structures. We
conclude that the proposed method is suitable for multimodal images with discrepancy structures.
Keywords Multimodal registration · Atlas · entropy · Discrepancy structures
1 Introduction
Multimodal medical imaging is an effective examination
method in medical diagnosis and computer-aided surgery
[14]. Multimodal images could provide physicians with
complementary information of patients’ tissues. However,
multimodal images usually need to be registered due to
different imaging resolutions and patients’ movement. The
registration of multimodal images is critical for effective
information fusion.
Various algorithms have been proposed for multimodal
image registration. Among those algorithms, maximizing
information entropy is a common type of multimodal reg-
istration methods. Various entropies are used as registration
measurements in these methods. Mutual information (MI) is
a widely used information entropy [15, 18]. It measures the
similarity between multimodal images by grayscale prob-
ability density. Normalized mutual information (NMI) is a
derivation of MI [20]. It could measure the image similarity
independent of registration images’ overlap region. Feature
neighborhood mutual information [12] and self-similarity
Houjin Chen
hjchen@bjtu.edu.cn
1
Beijing Jiaotong Unversity, Beijing, China
weighted mutual information [17] bring spatial informa-
tion to improve the registration performance. They intro-
duce the spatial information to resolve the complex texture
disturbance. Cross-cumulative residual entropy (CCRE) is
another information entropy to measure similarity between
multimodal images [22, 23]. CCRE is more robust as the
probability distribution is used. Those entropy-based meth-
ods focus on the gray correlation in multimodal images.
However, some structures in multimodal images are dis-
crepant. As shown in Fig. 1, the skull in Fig. 1aisabsent
compared with tissues in Fig. 1c. The discrepancy structure
changes the distribution of the entropy, which degrades the
entropy-based methods’ performance.
To deal with the structure discrepancy, we tried to
introduce atlases for multimodal registration. Atlas is a
widely used material in registration and segmentation [5, 10,
13]. Grouped by the type of the registration method, there
are two kinds of atlas-based registration approaches: feature
point-based methods [3] and intensity-based methods [7,
16]. Feature point-based methods match the feature points
extracted from images and atlases. Then a whole graph
transformation is applied based on the corresponding points.
Intensity-based methods define an objective function based
on images’ and atlases’ intensity. Then, the parameters
are optimized to maximize the objective function. Feature
point extraction is not needed in intensity-based methods.
According to the atlas’s type, the methods could be