Collaborative Research Aimed at Assisting Radiologists Presented at RSNA 2004

Researchers at DePaul University’s School of Computer Science, Telecommunications and Information Systems (CTI), working in collaboration with Northwestern University’s Feinberg School of Medicine, have developed a computer program designed to automatically classify tissue by organ or region of the body through analysis of a computed tomography (CT) scan image.

Preliminary results from an initial study, being presented this week at the Radiological Society of North America (RSNA) 2004 Scientific Assembly and Annual Meeting at Chicago’s McCormick Place, show that the new program has the potential to save the time and resources of radiologists and assist them in processing large volumes of patient data more rapidly.

In the study, the program was created using data from 344 CT images (collected and provided by Northwestern researchers) of the heart, liver, backbone, renal tissue and splenic parenchyma. DePaul researchers then implemented a segmentation algorithm using active contour maps that perform segmentation based on initial points on the boundary of the images and five main parameters that characterize the boundary of the sample. Then, 21 texture descriptors were calculated for each segmented region. Texture descriptors measure the variation of the intensity of a surface, and in a CT image, they measure a connected set of pixels satisfying a given gray level property which occurs repeatedly in an image region. Finally, a data mining algorithm based on decision trees was applied to automatically "learn" the relationships between different types of tissues and their texture descriptors.

The testing data demonstrates that the computer using the algorithm is able to identify CT tissue samples from the backbone with an accuracy rate of 98.6 percent, liver tissue samples at 92.5 percent accuracy, heart tissue at 93.2 percent accuracy, renal tissue at 96 percent accuracy, and the splenic parenchyma at 92.5 percent accuracy. Specificity for all of the different organ samples was higher than 95 percent, and for backbone, heart and renal tissue samples, specificity exceeded 97 percent.

These promising results support the conclusion that incorporation of additional texture models and descriptors will increase the performance of the classification program and can extend the functionality of the classifier to other organs and tissue regions.

"This shows us that we have the capability to pursue a larger data set, which will open up the possibility of including other organs and publishing a database of standard tissue characteristics to increase functionality," said Jacob Furst, associate professor at DePaul CTI and one of the study’s principal investigators, along with CTI assistant professor Daniela S. Raicu.

Data will be presented at RSNA 2004 by lead investigator Dr. David Channin, associate professor and chief of imaging informatics at Northwestern. The scientific poster session will be available in the InfoRAD technology exhibit area, Booth No. 9416-IMAi, located in McCormick Place’s Lakeside Center, Level 3, Hall D-2.

DePaul CTI is one of the most innovative and wide-ranging computer science programs in the country. The undergraduate program enrolls 1,200 students and offers 11 different degrees. More than 2,130 students are enrolled in its 17 graduate programs. CTI also features a doctoral degree program in computer science. With a total enrollment of 23,570 students on two city and five suburban campuses, DePaul is the seventh-largest private university in the U.S. and the largest Catholic university in the nation.

The Feinberg School of Medicine is a part of the vibrant educational and cultural community of Northwestern University, an independent private institution founded in 1851. The Feinberg School—one of Northwestern’s 11 colleges and schools—stands out among the nation’s medical schools. Consistently receiving high marks in U.S. News & World Report surveys, the medical school attracts bright and talented individuals to its faculty and student body.