{"title":"Clasificación automática de nódulos mamográficos basada en fusión de información visual multi-vista","authors":"Fabián Narváez","doi":"10.7476/9789978104910.0009","DOIUrl":null,"url":null,"abstract":"Correct mammography assessment and interpretation demands great expertise of radiologist observer and depends directly on a suitable visual analysis of mammographic findings and their correlation with radiographic features extracted from different mammographic views. In this paper, an automatic classification scheme for mammographic nodules contained on Regions of Interest (ROIs) is presented, which is based on an information fusion approach by using RoIs extracted from two different mammographic views of the same breast, a Mediolateral Oblique (MLO) view and a craniocaudal (CC) view, respectively. Once the expert radiologist selects a RoI from the two mammographic projections, those are characterized by using a multiresolution and multiscale decomposition approaches. For doing so, each RoI is projected into two different spaces defined by Zernike moments and Curvelet transform, respectively. Finally, this extracted heterogeneous information is optimally fused by using a Multiple Kernel Learning strategy based on Support vector machine scheme. The performance of the herein proposed strategy, for classifying benign and malignant nodules, was evaluated respect to the classical mammographic analysis based on only mammographic view, for which a set of 980 ROIs extracted from 490 cases and other set of 216 RoI extracted from 108 cases, which were extracted from DDSM and INBreast databases, respectively. The obtained results reported a sensitivity of 98.3% and a specificity of 94.5% respect to 96.2% and 93.1% of sensibility and specificity, respectively, and obtained by the analysis based on an only mammographic view. These results suggest that the herein proposed strategy could be useful in real clinic scenarios and could be contributing to the training of new radiologists.","PeriodicalId":319580,"journal":{"name":"Aplicaciones e innovación de la ingeniería en ciencia y tecnología","volume":"3 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Aplicaciones e innovación de la ingeniería en ciencia y tecnología","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.7476/9789978104910.0009","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Correct mammography assessment and interpretation demands great expertise of radiologist observer and depends directly on a suitable visual analysis of mammographic findings and their correlation with radiographic features extracted from different mammographic views. In this paper, an automatic classification scheme for mammographic nodules contained on Regions of Interest (ROIs) is presented, which is based on an information fusion approach by using RoIs extracted from two different mammographic views of the same breast, a Mediolateral Oblique (MLO) view and a craniocaudal (CC) view, respectively. Once the expert radiologist selects a RoI from the two mammographic projections, those are characterized by using a multiresolution and multiscale decomposition approaches. For doing so, each RoI is projected into two different spaces defined by Zernike moments and Curvelet transform, respectively. Finally, this extracted heterogeneous information is optimally fused by using a Multiple Kernel Learning strategy based on Support vector machine scheme. The performance of the herein proposed strategy, for classifying benign and malignant nodules, was evaluated respect to the classical mammographic analysis based on only mammographic view, for which a set of 980 ROIs extracted from 490 cases and other set of 216 RoI extracted from 108 cases, which were extracted from DDSM and INBreast databases, respectively. The obtained results reported a sensitivity of 98.3% and a specificity of 94.5% respect to 96.2% and 93.1% of sensibility and specificity, respectively, and obtained by the analysis based on an only mammographic view. These results suggest that the herein proposed strategy could be useful in real clinic scenarios and could be contributing to the training of new radiologists.