Computer Assisted Diagnose of Intra Pleural Lung Nodules using Non-Parametric Classifiers

Andre Hallack and Joceli Mayer


Lung nodule CAD, Medical imaging, Image segmentation, Pattern recognition


This work proposes a new methodology for computer assisted diagnose of intra pleural lung nodules. The nodules candidates are initially selected based on shape index and the segmentation is based on a proposed region growing algorithm. The proposed system may employ a variety of non-parametric classifiers, where the best one is selected by a performance analysis. The amount of features for the classification step was properly trimmed using the Sequential Floating Backward Selection algorithm. The performance of the proposed system is trained and assessed through 110 annotated series from the Lung Image Database Consortium for various classifiers, presenting a 73\% sensitivity performance with only 2.12 false positives per exam. Moreover, an independent evaluation is done through the CAD lung nodule competition ANODE09 ranking it the third best system.

Important Links:

Go Back