Automated marker identification using the Radon transform for watershed segmentation. Issue 3 (1st March 2017)

• Record Type: Journal Article
• Authors: González‐Betancourt, Aniel; Rodríguez‐Ribalta, Patricia; Meneses‐Marcel, Alfredo; Sifontes‐Rodríguez, Sergio; Lorenzo‐Ginori, Juan Valentín; Orozco‐Morales, Rubén
• Is Part Of: IET image processing. Volume 11:Issue 3(2017)
• Page Start: 183
• Page End: 189
• Publication Date: 2017-03-01
• Subjects: matched filters -- image filtering -- Radon transforms -- image segmentation -- medical image processing -- optical microscopy; automated marker identification -- Radon transform -- watershed segmentation -- medical imaging -- digital image analysis -- low computational cost system -- grey‐scale image circular structure edge -- red blood cells -- optical light microscopy -- blood smears -- shape parameter detection -- sinogram projections -- matched filter; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/iet-ipr.2016.0525
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract : The automated identification of markers in medical imaging is one of the issues proposed in recent years for many tasks with wide application in digital image analysis, but there is no universal algorithm for this. We introduce here a low computational cost system based on the Radon transform (RT) to determine markers for watershed segmentation purposes in nearly circular structures of grey‐scale images such as the red blood cells in images obtained by optical light microscopy of blood smears. The search for markers is performed based on the ability of the RT to detect shape parameters and characterise their behaviour in circular structures. The circular structure edge was determined previously in order to apply the direct RT and after that the sinogram projections were filtered utilising a matched filter. Then, an image with peaks close to the circular structures' centres is obtained by means of the reverse RT. After this, a threshold is calculated and applied in order to identify the markers. The reached precision was 97.4%, which is comparable with results within the state of the art, and the F‐measure was 0.925.