Classification of SD-OCT volumes with multi pyramids, LBP and HOG descriptors: application to DME detections

Khaled Alsaih (1) , Guillaume Lemaître (1) , Joan Massich Vall (1) , Mojdeh Rastgoo (1) , Désiré Sidibé (1) , Tien y Wong (2) , Ecosse Lamoureux (2) , Dan Milea (2) , Carol Y. Cheung (2) , Fabrice Mériaudeau (1)

1 Le2i - Laboratoire Electronique, Informatique et Image [UMR6306]
2 Singapore Eye Research Institute, Singapore National Eye Centre

This paper deals with the automated detection of DME on OCT volumes. Our method considers a generic classification pipeline with preprocessing for noise removal and flattening of each B-Scan. Features such as HoG and LBP are extracted and combined to create a set of different feature vectors which are fed to a linear-SVM classifier. Experimental results show a promising sensitivity/specificity of 0.75/0.87 on a challenging dataset.

Domaines: Sciences de l'ingénieur [physics]; Traitement du signal et de l'image [eess.SP]; Informatique [cs]; Imagerie médicale; Informatique [cs]; Vision par ordinateur et reconnaissance de formes [cs.CV]

Khaled Alsaih, Guillaume Lemaître, Joan Massich Vall, Mojdeh Rastgoo, Désiré Sidibé, et al.. Classification of SD-OCT volumes with multi pyramids, LBP and HOG descriptors: application to DME detections. 38th IEEE Engineering in Medicine and Biology Society (EMBC), Aug 2016, Orlando, United States. ⟨hal-01320212⟩