A University of Rochester research team sought to apply new methods for texture analysis, designed to classify unclear tissue lesions based on the spatial pattern of their appearance on images, said presenter Mahesh Nagarajan, a graduate student at the Rochester, NY-based institution.
While the results are preliminary, the researchers will show that advanced texture analysis methods demonstrate considerable improvement in characterizing breast lesions over traditional approaches, he said. The research also highlights the potential role for case-based image retrieval aimed for classifying suspicious breast lesions.
"Rather than simply consider the morphology of disease patterns in the study being read, such a neural network-based image retrieval approach would also incorporate information from previously read cases while assisting a radiologist with the current case," Nagarajan told AuntMinnie.com. "This could very well serve as the next major improvement in CAD applications and possibly addresses some of the current concerns affecting CAD such as the high rate of false positives."
