-
Categories
-
Pharmaceutical Intermediates
-
Active Pharmaceutical Ingredients
-
Food Additives
- Industrial Coatings
- Agrochemicals
- Dyes and Pigments
- Surfactant
- Flavors and Fragrances
- Chemical Reagents
- Catalyst and Auxiliary
- Natural Products
- Inorganic Chemistry
-
Organic Chemistry
-
Biochemical Engineering
- Analytical Chemistry
-
Cosmetic Ingredient
- Water Treatment Chemical
-
Pharmaceutical Intermediates
Promotion
ECHEMI Mall
Wholesale
Weekly Price
Exhibition
News
-
Trade Service
Radiation therapy is a technology that uses radiation to eliminate tumor cells at a targeted point, which is an important technical means for
cancer treatment.
In order to maximize the irradiation of tumor lesions while protecting surrounding tissues and organs, multimodal imaging (computed tomography (CT), magnetic resonance (MRI), ultrasound (US), and conical beam CT (CBCT)) is based ) and other guided radiotherapy techniques have received great attention
.
Among them, cone beam CT (CBCT) images have the advantages of high bone tissue contrast and high spatial resolution, and CBCT image-guided radiotherapy is the most widely used image guidance technology
compared with other image guidance technologies.
Precise radiotherapy of tumors is made possible
by rigid or elastic registration of positioning CT images and CBCT images scanned during the treatment implementation stage, and positioning and dose verification between fractional treatments.
However, due to the grayscale difference between CT and CBCT images, the inconsistency of structural information, the poor quality of CBCT images and other factors, fast and accurate CT-to-CBCT The study of image registration algorithms is still very challenging (as shown in Figure 1, there are more serious artifacts
in CBCT images at the same anatomical position of CBCT and CT).
Traditional registration algorithms generally use iterative optimization algorithms, which have long running time and poor
real-time performance.
At present, the frontier of related research work mainly focuses on the use of deep learning theory to study fast and accurate registration methods
.
However, these efforts have not been studied in depth
in the face of the distribution difference between the CBCT and CT image domains, as well as noise artifact interference in CBCT.
Figure 1.
Images at the same anatomical location as CBCT and CT
In view of this problem, Yang Xiaodong's research group of Suzhou Medical Institute proposed a registration algorithm
based on boundary gradient guidance and cross-domain feature fusion.
The overall structure of the algorithm is shown in Figure 2, which contains two important modules: boundary-guided attention module (EGAM) and cross-domain attention module (CDAM), which together form a cross-domain converged registration network
.
The network uses two convolutional streams with the same structure to extract the unique image features
in the two image domains of CT and CBCT in an uncoupled manner.
In addition, the boundary-guided attention module fully mines the boundary information of gradient images, guides the registration network to model the correspondence between relevant anatomical structures in CT and CBCT, and suppresses noise artifacts in CBCT.
The cross-domain attention module uses global and local information to guide features from two image domains to map to a common space to mitigate distribution differences
between image domains.
The algorithm is experimentally performed on real clinical CT-CBCT datasets and achieves the best performance
compared with other advanced registration methods.
Compared with the traditional registration method, the method has significantly improved
the TRE, DSC, and MHD indicators.
Among them, the TRE error was reduced from 4.
00mm to 2.
27mm, and the DSC index was increased from 74.
02%.
80.
01%, MHD distance also reduced from 1.
62mm to 1.
50mm
.
Under the same hardware conditions, the method has a nearly 10-fold increase
in operating speed.
In addition, the algorithm also achieves competitive registration performance on the public lung 4D-CT dataset (Dir-Lab), demonstrating the potential
of the proposed method in single-mode image registration.
In the future, the team will conduct more in-depth research on the pain points of multimodal image registration in IGRT to help improve
the accuracy and efficacy of clinical radiotherapy.
Figure 2.
( a) CDFRegNet network framework; (b) EGAM module network structure; (c) CDAM module network structure
Table 1 Quantitative results of different methods
Figure 3 Visualization of different registration methods
The research result "CDFRegNet: A Cross-domain Fusion Registration Network for CT-to-CBCT Image Registration.
" Published in
the journal Computer Methods and Programs in Biomedicine.
The first author of the paper is Cao Yuzhu, a graduate student, and Professor Zheng Jian of Suzhou Medical Engineering Institute and Professor Ni Xinye of Changzhou Second Hospital are the corresponding authors
.
This work has been supported by the Natural Science Foundation of Shandong Province (ZR2021MH213), Suzhou Science and Technology Bureau (SS202087, SJC2021023), Jiangsu Provincial Health Commission ( M2020006), Changzhou Key Laboratory of Medical Physics (CM20193005) and other projects
.
Links to papers:
