This work is a part of the MAGIC-5 (Medical Applications on a Grid Infrastructure Connection) experiment of the Italian INFN (Istituto Nazionale di Fisica Nucleare). A simple CAD (Computer-Assisted Detection) system for juxta-pleural lung nodules in CT images is presented, with the purpose of comparing different 2D concavity-patching techniques and assessing the respective efficiency in locating nodules. After a short introduction on the motivation, and a review of some CAD systems for lung nodules already published by the MAGIC-5 Collaboration, the paper describes the main lines of this particular approach, giving preliminary results and comments. In our procedure, candidate nodules are identified by patching lung border concavities in a hierarchical multiscale framework. Once located, they are fed to an artificial neural network for false positive reduction. The system has a modular structure that easily allows the insertion of arbitrary border-smoothing functions for concavity detection and nodule searching. In this paper the a-hull and morphological closing are compared, proving the higher sensitivity of the former, which also appears computationally less heavy. (C) 2010 Elsevier B.V. All rights reserved.
Approaches to juxta-pleural nodule detection in CT images within the MAGIC-5 Collaboration
FANTACCI, MARIA EVELINA;
2011-01-01
Abstract
This work is a part of the MAGIC-5 (Medical Applications on a Grid Infrastructure Connection) experiment of the Italian INFN (Istituto Nazionale di Fisica Nucleare). A simple CAD (Computer-Assisted Detection) system for juxta-pleural lung nodules in CT images is presented, with the purpose of comparing different 2D concavity-patching techniques and assessing the respective efficiency in locating nodules. After a short introduction on the motivation, and a review of some CAD systems for lung nodules already published by the MAGIC-5 Collaboration, the paper describes the main lines of this particular approach, giving preliminary results and comments. In our procedure, candidate nodules are identified by patching lung border concavities in a hierarchical multiscale framework. Once located, they are fed to an artificial neural network for false positive reduction. The system has a modular structure that easily allows the insertion of arbitrary border-smoothing functions for concavity detection and nodule searching. In this paper the a-hull and morphological closing are compared, proving the higher sensitivity of the former, which also appears computationally less heavy. (C) 2010 Elsevier B.V. All rights reserved.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.