Automatic lesion detection and volume measurement in MR imaging of plexiform neurofibromas

Conventional response criteria (RECIST, WHO) are inadequate for plexiform neurofibromas (IN) due to their size and complex shapes. An automated method was developed to detect and quantify the volume of PN on STIR MR images. The automated algorithm implements heuristics derived from human-based recognition of lesions (pixel intensity contrast and edge detection/following). A connected component analysis distinguishes multiple non-contiguous lesions and removes lesions considered too small, and an edge following algorithm defines the border of the lesion. This method was validated by two observers, who performed automated volume calculations and manual tracings to estimate tumor volume. The method was reproducible (C.V., 0.6% to 5.6%), and the inter-observer difference in the average tumor volume ranged from 6% /spl plusmn/ 3.8 to -5.2% /spl plusmn/ 5.4. The automated and manual methods of volume determination yielded similar results (R = 0.999). The automated method will likely improve the reproducibility and sensitivity of response assessment in clinical trials for patients with PN.