17 November 2017 Scoring nuclear pleomorphism using a visual BoF modulated by a graph structure
Author Affiliations +
Proceedings Volume 10572, 13th International Conference on Medical Information Processing and Analysis; 1057205 (2017) https://doi.org/10.1117/12.2286007
Event: 13th International Symposium on Medical Information Processing and Analysis, 2017, San Andres Island, Colombia
Abstract
Nuclear pleomorphism has been recognized as a key histological criterium in breast cancer grading systems (such as Bloom Richardson and Nothingham grading systems). However, the nuclear pleomorphism assessment is subjective and presents high inter-reader variability. Automatic algorithms might facilitate quantitative estimation of nuclear variations in shape and size. Nevertheless, the automatic segmentation of the nuclei is difficult and still and open research problem. This paper presents a method using a bag of multi-scale visual features, modulated by a graph structure, to grade nuclei in breast cancer microscopical fields. This strategy constructs hematoxylin-eosin image patches, each containing a nucleus that is represented by a set of visual words in the BoF. The contribution of each visual word is computed by examining the visual words in an associated graph built when projecting the multi-dimensional BoF to a bi-dimensional plane where local relationships are conserved. The methodology was evaluated using 14 breast cancer cases of the Cancer Genome Atlas database. From these cases, a set of 134 microscopical fields was extracted, and under a leave-one-out validation scheme, an average F-score of 0.68 was obtained.
© (2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Ricardo Moncayo-Martínez, Ricardo Moncayo-Martínez, David Romo-Bucheli, David Romo-Bucheli, Viviana Arias, Viviana Arias, Eduardo Romero, Eduardo Romero, } "Scoring nuclear pleomorphism using a visual BoF modulated by a graph structure", Proc. SPIE 10572, 13th International Conference on Medical Information Processing and Analysis, 1057205 (17 November 2017); doi: 10.1117/12.2286007; https://doi.org/10.1117/12.2286007
PROCEEDINGS
10 PAGES


SHARE
Back to Top