Landmark Detection in the Chest and
Registration of Lung Surfaces in
Computed Tomography Scans





Background


Lung cancer remains the leading cause of cancer death in the US.  The overall 5-year survival rate is only 15%, but early detection and resection of pulmonary nodules in Stage I can improve the prognosis to 67%.  The curability of early stage lung cancer has motivated researchers to propose lung cancer screening.  Repeated computed tomography (CT) scans are needed in order for the radiologist to determine whether a small nodule is growing.  Since these scans contain hundreds of images, it is time consuming for the radiologist to manually follow up nodules.


Methods


Our automated system registers CT images temporally.  The system detects anatomical landmarks, in particular, the trachea, sternum, and spine, using an attenuation-based template matching approach.  It computes the optimal rigid-body transformation that aligns the corresponding landmarks in two CT scans of the same patient.  This transformation then provides an initial registration of the lung surfaces segmented from the two scans.  The initial surface alignment is refined step by step in an iterative closest-point (ICP) process.  The surface transformation is applied to align nodules in the initial CT scan with nodules in the follow-up scan.

 

Landmark Detection

 

    Trachea  

 

 

    Sternum 

 

 

    Vertebra    

 

Figure above is coronal view of a chest CT scan. The yellow line marks the most cranial image with visible lung (A), the purple line the axial image at the carina (B). The trachea in image A (green) is detected by correlation-based template matching using the generic trachea template on the left. Sternum and vertebra in image B (light and dark blue) are detected using their respective generic templates. The trachea in image B (green) is found using a template cropped online from the preceding axial image.

 

Initial Landmark Registration

 

Four points used for registration are shown for each scan: the center of the trachea cross-section in slice A and the centers of the cross-sections of sternum, trachea, and vertebra in slice B in each study. The landmarks in study 1 (green) are then being matched to the landmarks in study 2 (red).

 

Iterative Closest Point (ICP)

              

3 Steps

1.            Find correspondence of surface points

2.            Minimize sum of squared errors:

 ∑ || pi  - R xi  - t ||2              (Closed-form solution by Horn 87)

  i

3.            If change is less than threshold, transform surface points x, then repeat steps 1-3             

 

Lung Surface Registration Result

 

Before Initial Registration

After Initial Registration

Registration after 25th Iterations of ICP


Here is sample registration result for 5mm thickness lung surfaces.  The lung surfaces are shown before initial registration, after initial registration and after 25 iteration of the iterative closest point algorithm (ICP).  The lungs in scan 1 are shown in red and in scan 2 prior to registration in green and after final registration in blue.  The registration process shifted the surfaces in scan 2 to the right and slightly rotated them to align with the surfaces in scan 1

 

Efficient Algorithms

 

 

These algorithms were used to improve efficiency as well as accuracy in our lung surface registration process.

 

Nodule Results

 

For 56 out of 58 nodules in the initial CT scans of 10 patients, nodule correspondences in the follow-up scans are established correctly.  Our methods, therefore, can potentially facilitate the radiologist's evaluation of pulmonary nodules on chest CT for interval growth.  Here is an example nodule correspondence below.

 

 

 

 

Registration results for 5mm thickness lung surfaces.  The lung surfaces are shown before and after registration.  The nodules in scan 1 that are registered are shown in red and nodules in scan 2 shown in blue.

 

Movies of Registration Process

 

Click on link to save movie file
 

 

 

 

 

 

 

Relevant papers:

 

H. Hong, M. Betke, S. Teng, D. Thomas, and J. P. Ko, "Multilevel 3D Registration of Lung Surfaces in Computed Tomography Scans -- Preliminary Experience," International Conference on Diagnostic Imaging and Analysis, ICDIA 2002, pp. 90-95, Shanghai, China, August 2002.

 

J. P. Ko, H. Rusinek, E. Jacobs, J. S. Babb, M. Betke, G. McGuinness, and D. P. Naidich, "Volume Measurement of Small Pulmonary Nodules on Chest CT: A Phantom Study," accepted by Radiology, April 2002.

 

J. P. Ko, H. Rusinek, E. Jacobs, R. Chandra, G. McGuinness, M. Betke, D. P. Naidich. "Volume quantitation of small pulmonary nodules on low dose chest CT: a phantom study." To be presented at the Radiological Society of North America 87th Scientific Assembly and Annual Meeting RSNA 2001, Chicago, IL, November 2001.

 

M. Betke, H. Hong, and J. P. Ko, "Automatic 3D Registration of Lung Surfaces in Computed Tomography Scans," Fourth International Conference on Medical Image Computing and Computer-Assisted Intervention MICCAI 2001, pp. 725--733, Utrecht, The Netherlands, October 2001.

 

J. P. Ko and M. Betke, "Chest CT: Automated Nodule Detection and Assessment of Change over Time-Preliminary Experience." Radiology, 218, 267-273, January 2001.

 

M. Betke and J. P. Ko, "Detection of Pulmonary Nodules on CT and Volumetric Assessment over Time." Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 245--252, Cambridge, UK, September 1999.