Scattering features for lung cancer detection in fibered confocal fluorescence microscopy images
Résumé
Fibered confocal fluorescence microscopy (FCFM) imaging technique is a novel medical imaging technique which interest is yet to be establish for diagnosis problem. This paper addresses the problem of lung cancer detection using FCFM images and as a first contribution, assesses the feasibility of computer aided diagnosis through these images. For doing so, we have built a pattern recognition scheme which involves a feature extraction and a classification stages. The second contribution relies on the used features for discrimination. Indeed, we have employed the so-called \emph{scattering transform} for extracting discriminative features robust to small deformations in the images. We have shown that these features yield to better recognition performances than classical yet powerful features like \emph{Local Binary Patterns} (LBP) for our FCFM image classification problems and are competitive to LBP on other medical imaging classification problems. Another of our findings is that LBP and scattering-based features provides complementary discriminative informations and in some situations, we have empirically established that enhanced performances can be obtained when jointly using LBP and scattering features.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...