Automatic detection and segmentation of abdominopelvic lymph nodes on computed tomography scans

We proposed a local scale-based Hessian analysis method for automated lymph node detection in contrast-enhanced abdominopelvic CT scans. First, spine and pelvic girdle were automatically segmented to locate the abdominopelvic region. Blood vessels were then segmented to narrow the search region to the perivascular space where lymph nodes are located. Lymph node candidates were generated by scale-based Hessian analysis. The detected candidates were segmented by curve evolution. Characteristic features were calculated on the segmented nodes. Support vector machine was utilized for classification and false positive reduction. We applied our method to 22 patients with 37 enlarged lymph nodes. The system achieved 83% sensitivity at 5 false positives per patient. Our results indicated that computer-aided abdominopelvic lymph node detection is feasible with a high sensitivity and a relatively low FP rate.