Segmenting mammographic microcalcifications using a semi-automatic procedure based on Otsu s method and morphological filters

• DOI: 10.4322/rbeb.2013.037
• Publication Date: 2013-11-14T12:16:56Z
• Authors: Marcelo de Almeid...; André Victor Alva...; Carolina Maria Az...; Maria Julia Gregó...; Antonio Fernando ...; Wagner Coelho de ...
• Keywords: Computer-Aided Diagnosis; Region of interest; Microcalcification

Introduction: Breast cancer has the second highest world's incidence rate, according to Brazilian National Cancer Institute (INCa).Clinical examination and mammography are best methods for early diagnosis.Computer-aided detection (CADe) computer-aided diagnosis (CADx) systems developed improve mammographic diagnosis.Basically, CADx have three components: (i) segmentation, (ii) parameters extraction selection, (iii) lesion classifi cation.The fi rst step a system is segmentation.Methods: A microcalcifi cation segmentation method proposed, based on morphological operators, Otsu's Method radiologists' knowledge.Pre-processing with top-hat operators improves contrast reduces background noise.The automatically selects grey-level threshold segment cations, obtaining binary images.Following, inferior reconstruction dilatation applied reconstruct lost structure details ll small fl aws in segmented cations.Finally, Canny edge identify cations contour candidates each region-of-interest (ROI).Two experienced radiologists intervene this semi-automatic method, rstly, selecting ROI and, then, analyzing result.The was assessed 1000 ROIs from 158 digital images (300 dpi, 8 bits).Results: Considering opinion, rates of adequately establish hypothesis were 97.8% one radiologist 97.3% other.Using Area Overlap Measure (AOM) 2136 delineated by an as gold standards, achieved average AOM 0.64±0.14,being 0.56±0.09for 0.66±0.13for large ones.Moreover, 0.64±0.13for benign 0.64±0.14for malignant lesions no statistical differences between them.Conclusion: Based these ndings, proposed could be used develop that help breast detection.