Skip to main content

Computer-Aided Detection of Epidural Masses in Computed Tomography using a Constrained Gaussian Mixture Model Framework

Thursday, November 07, 2013 — Poster Session II

12:00 p.m. – 2:00 p.m.

FAES Academic Center (Upper-Level Terrace)




  • s Pattanaik
  • j Liu
  • J Yao
  • W Zhang
  • EB Turkbey
  • RM Summers


PURPOSE: Our preliminary Computer-Aided Detection (CAD) framework aims to localize candidate epidural mass detections and reduce false positives. MATERIALS AND METHODS: 23 patients with chest-abdomen-pelvis CT with confirmed epidural masses were selected. 17 patients without mass were served as controls. Two radiologists manually demarcated the ground truth. The CAD system segments whole spine using a watershed algorithm and directed graph search. It also isolates the spinal canal. Four tissue classes were generated using K-means clustering to represent normal intradural tissue, fat/vasculature, the epidural mass, and a partial volume region between the bone and soft tissue. CGMM was employed to refine classification, taking advantage of both spatial and intensity parameters. Detections were limited to masses extending from the canal boundary. RESULTS: Before classification with SVM, our CAD system detected 44 out of 47 detections. A sensitivity of 80% with 7.2 false positives per patient was attained following classification and ten-fold cross-validation, which compared favorably against the sensitivity of 76% with 7.4 false positives per patient attained by restricting CAD to intensity based K-Means clustering. CONCLUSION: Our CAD system lays the groundwork for detection of epidural masses in CT using a combination of spatial and intensity based parameters to localize epidural masses.

back to top