Skip to main content

Detecting retroperitoneal lymphadenopathy on CT scans using multi-cue kernel density estimators

Thursday, November 07, 2013 — Poster Session II

12:00 p.m. – 2:00 p.m.

FAES Academic Center (Upper-Level Terrace)




  • K.M. Cherry
  • S. Wang
  • R.M. Summers


Lymphadenopathy analysis is an important part of daily clinical work in radiology. Automatic detection of enlarged lymph nodes in CT scans is a very challenging task due to variations of location and shape of the nodes, especially in the abdominal region. In this work, we propose a computer-aided detection (CAD) system for enlarged abdominal lymph nodes on CT scans based on multiple feature cues. The fully-automated method constructed probability density functions using kernel density estimation for each feature computed on the lymph nodes in the training set. The probability that each voxel belongs to a lymph node was calculated from the trained functions, and local probability maximums were used as potential lymph node candidates. We extracted shape, texture, and intensity features from each candidate and employed support vector machines for classification. Preliminary experimental results showed that the proposed CAD performed well on a data set with 48 patients. The system detected 76% of the lymph nodes with 15.4 false positives per patient.

back to top