Metastatic cancer early detection on lung and brain
Investigators: W. O'Dell, Robert Ambrosini, Peng Wang (BME PhD '08).

Aims
Novel applications of conformal stereotactic radiation therapy championed by our clinical colleagues have renewed the clinical motivation for early detection of metastatic cancer to brain and lung. The major challenge in the lung is that the small nodules of interest resemble in a 2D slice the numerous blood vessels coursing through the lung tissue. In the brain one has to deal with differential uptake of contrast in T1-weighted post-contrast images. We have developed and patented a novel 3D template matching approach that seem to far outperform any existing commercial CAD system or any other method presented in the literature.

Figure 1 [A] CT image slice through the chest of a patient with 5 lung mets, 2 of which are apparent at this slice location (arrows).	[B] Schematic of one of our 3D nodule appearance models (aka 'templates'). The variable intensities are at pixel ar due to partial volume effects at this image resolution, slice thickness, and the dimensions of the model.	[C] Map of the correlation values of our 3D templates to the image features in this 3D CT dataset, but shown only on this slice. The 2 brightest pxiels correspond to the 2 real tumors.

Figure 2. Representative T1-weighted MR images of patients with mets to the brain.

Future Work
Detection of nodules in the body is not equivalent to a diagnosis of metastatic disease. For that, accurate quantification of tumor growth over time from sequential images is needed. Expert radiologists are notoriously poor at estimating tumor volumes, as evidenced by the results of the recent Lung Image Database Consortium (LIDC) multi-institutional study. We have been working on an automated method for computing tumor volumes, and thereby growth, using the matching features of our 3D template algorithm. This project is described in detail on our Tumor Volume Estimation page.

We have applied this approach successfully to images of a breast phantom scanned using a new breast-dedicated cone-beam CT scanner, and less successfully to detection of lymph nodes in the upper chest and neck. We hope to expand this method to the detection of tumors of the liver and colon polyps a part of a virtual colonoscopy scheme.

Related Publications

1. Wang P., DeNunzio A., Okunieff P., O’Dell W.G.
   Metastatic lung tumor early detection using 3D template matching
         Medical Physics, (March) 34(3):915-922, 2007
2. Ambrosini R, Wang P, O'Dell WG
   Volume change determination of metastatic lung tumors in CT images using 3-D template matching.
   Proceedings of SPIE Medical Imaging [#7260-112], Orlando, FL, Feb 2009
3. Ambrosini R, Wang P, O'Dell WG
   Computer-Aided Detection of Metastatic Brain Tumors Using Automated 3-D Template Matching
   (submitted to the journal of Medical Physics, March 2009)

Related Presentions

1. Ambrosini R, Wang P, Victor J, Sobe N, O'Dell WG
   Automatic Detection and Sizing of Metastatic Brain Tumors Using 3D Template Matching
   47th Annual AAPM Meeting (SU-FF-I-2), Seattle WA, July 2005