Namkug Kim, PhD, and colleagues from the University of Ulsan developed the semiautomated CAD scheme and applied it to images of 59 patients, including 30 biopsy-proven cases. The CAD algorithm was designed to extract image features based on nodule segmentation to perform automated classifications.
The algorithm yielded sensitivity of about 95% with specificity of 93%, a level of accuracy that is equivalent to a radiologist's interpretation, according to the group.
Still, the technique needs refinement. The semiautomated CAD system needs radiologists to select the center and edge point of the lesion. A larger sample size is also needed to validate the approach.
"Even though this study has several limitations in that the CAD is semiautomatic, it requires radiologists to select an approximate center and edge point for generating the initial [region of interest] in the segmentation process, and the sample size is small, fully automated CAD with refined segmentation could be developed and used as a second opinion in clinical circumstances," Kim told AuntMinnie.com.
