Medical imaging series are acquired in many formats using different technologies and equipment, depending on the anatomy or function of interest. In some cases an imaging series may not provide enough information on its own and may benefit from being reviewed concurrently with another imaging series. For accurate comparison the two imaging series will need to be aligned, because patients are often in different positions from one imaging series to the next. The process of aligning imaging series is called image registration. Registered series are capable of displaying the data of each individual scan in the same relative space, facilitating a more comprehensive understanding of anatomical relationships and functions.
Figure A (Above): CT and MR imaging data (left) are unaligned before registration. Aligned data (right) shows the correct anatomical positioning after registration.
How is Image Registration Performed?
The process of image registration typically involves these steps:
• Image Acquisition: Medical images are acquired via different modalities, such as CT, MRI, PET, ultrasound, etc.
• Pre-processing: Pre-processing steps may be performed before registration to reduce noise, enhance the images, or highlight specific anatomy through segmentation.
• Feature Identification: Key features are identified from the images such as edges, corners, or other landmarks that are visible in both datasets.
• Registration Algorithm: Various registration algorithms can be used to align the images:
• Feature-based registration: Involves identifying common features, such as corners or edges, then rotating the dataset so the features become aligned to the other dataset.
• Intensity-based registration: Compares the intensity values of the Hounsfield Units (HU) of the two datasets and finds a rotation that minimizes the difference between the two.
• Hybrid registration: Combines both feature-based and intensity-based registration to improve accuracy.
• Validation: The registered images are reviewed for accuracy using metrics that evaluate the relationship between two variables (correlation coefficient). This can be done using software or algorithms, and usually requires input and interpretation from medical professionals.
• Post-processing: The registered images may require further processing such as segmentation to extract useful information.
Figure B (Above): Visual workflow for image registration.
Medical image registration can be a critical tool in healthcare aiding in diagnosis, treatment planning, and treatment monitoring. Aligned scans are useful in research as well when studying disease progression, treatment outcomes, and other important factors. The 3DQ Lab uses registration for several 3D services including Tumor Quantification, Functional MRI, Epilepsy Mapping, Neurosurgical Navigation, and more.
3DQ Lab – Grant Building
300 Pasteur Drive
Stanford, CA 94305
Directions to Grant
3DQ Lab – Clark Building
318 Campus Drive
Stanford, CA 94305
Directions to Clark
Learn More About the Lab
Copyright © Stanford University