Wavelet-Based Multiscale Texture Segmentation: Application to Stromal Compartment Characterization on Virtual Slides.

Nicolas Signolle (1, 2) , Marinette Revenu (1) , Benoît Plancoulaine (2) , Paulette Herlin (2)

1 Equipe Image - Laboratoire GREYC - UMR6072
2 GRECAN - Groupe Régional d'Etudes sur le CANcer

We aim at segmenting very large images of histopathology virtual slides with an heterogeneous and complex content. To this end, we propose a multiscale framework for texture-based color image segmentation. The core of the method is based on a wavelet-domain hidden Markov tree model and a pairwise classifiers design and selection. The classifier selection is founded on a study of the influence of the hyperparameters of the method used. Over the testing set, majority vote was found to be the best way of combining outputs of the selected classifiers. The method is applied to the segmentation of various types of ovarian carcinoma stroma, on very large virtual slides. This is the first time such a segmentation is tested. The segmentation results are presented and discussed.

Mots clés: Segmentation; Texture; Hidden Markov Model; Wavelets; Pairwise classifier combination; Virtual slides

Domaines: Traitement des images [eess.IV]; Imagerie médicale

Nicolas Signolle, Marinette Revenu, Benoît Plancoulaine, Paulette Herlin. Wavelet-Based Multiscale Texture Segmentation: Application to Stromal Compartment Characterization on Virtual Slides.. Signal Processing, 2010, 90 (8), pp.2412-2422. ⟨10.1016/j.sigpro.2009.11.008⟩. ⟨hal-00805756⟩