Lymph-nodes automatic detection lung-cancer Hessian Harris Region Growing Mediastinal
Abstract:
The central-chest lymph-node biopsy is a major diagnostic procedure that greatly influences the efficacy of lung-cancer staging. Finding the accurate locations of signifi cant lymph nodes is a vital step in the diagnosis of lung cancer. In this thesis, we aim to develop a time-efficient and robust method for automatically detecting central-chest lymph-nodes in 3D MDCT chest images. The method employed here models lymph-nodes as 3D-ellipsoidal blob structures. The lymph-nodes are detected by a Harris-based Hessian analysis on the mediastinum region by removing the presegmented structures of the central-chest region followed by a customized region-growing and false-positive reduction. The method achieves a true detection rate of 57% and 61% with an average of 47 and 48 false-positives per case for lymph-nodes with short-axis length >= 7 mm and >= 10 mm respectively, evaluated over a dataset of 17 cases. The average computation time for the method is 31 minutes.
