Their study appeared in the March issue of Pattern Recognition.
Lung cancer is the deadliest cancer in men and women. According to the National Institutes of Health, the five-year survival rate (16.3 percent) is worse than many other cancers, such as colon (65.2 percent), breast (90.0 percent) and prostate (99.9 percent). More accurate tumor imaging, in terms of tumor feature extraction, could improve diagnostic and predictive accuracy
“The new method we developed will improve diagnostic accuracy and make more individualized cancer care possible,” said study senior author Robert J. Gillies, Ph.D., chair of the Department of Cancer Imaging and Metabolism at Moffitt. “It will improve our ability to quantify changes in cancer and respond appropriately with therapy.”
Tumor segmentation was previously a difficult task because of the diverse composition of cancer lesions when compared to normal tissues. The new segmentation method marks a great improvement over a previously used manual method, said the researchers.
“A common approach to delineate lung cancer tumors is for the radiologist or radiation oncologist to manually draw the boundary of the tumor,” explained Gillies. “This method is variable and operator-dependent. A highly automatic, accurate and reproducible lung tumor delineation algorithm would offer a significant advance.”
Their development of SCES offers that advancement, and because the process is automated, it requires less time and effort.
“A big advantage with single click ensemble segmentation is that it only requires one human interaction – the manual seed input. This is when the radiologist or radiation oncologist places the seed points in the tumor area,” Gillies said. “With SCES, lesion delineation was accurate and consistent, and the lung segmentations workload was greatly reduced.”
The new algorithm uses the original by incorporating the original seed point to define an area within which multiple seed points are automatically generated. Ensemble segmentation can then be obtained from the multiple regions.
According to the researchers, the measurement can be used to determine if the tumor is increasing or decreasing in size, as well as describe features such as shape and texture.
“With this method, all the radiologist has to do is click their mouse on a tumor and the program will automatically perform an accurate measurement,” explained Hall. “We also demonstrated that this approach reduces inter-observer variability with significantly fewer operator interactions when compared with the original algorithm.”
The capabilities of the new algorithm were successfully tested on a large patient tumor imaging data set.
Their work was funded by National Institutes of Health Grant 1U01 CA 143062-01
source : http://www.sciencedaily.com/releases/2013/05/130502115527.htm