A Review on Automatic Mammographic Density and Parenchymal Segmentation

Table 2

Summary of representative studies using clustering based methods for mammographic tissue segmentation. denotes correlation coefficient. Note that () largely identical studies are excluded in the list and () in case of multiple results, only the best reported results are listed.

Study	Year	Number of density categories	Modalities	Number of views	Number of images	Segmentation evaluation	Risk/density estimation accuracy
General clustering
Oliver et al. [38]	2005	Fatty and dense	Digitised SFM	MLO	180 (MIAS)	N/A	73% (fatty, glandular, and dense)
Strange et al. [39]	2013	Dense and fatty	FFDM	CC	12	Visually assessed	N/A
Marias et al. [40]	2005	Dense, semidense, and fatty	Digitised SFM	MLO	146	Visually assessed	65% (BI-RADS I and II); 86% (BI-RADS III and IV)
Oliver et al. [41]	2005	Dense and fatty	Digitised SFM	MLO	300 (DDSM)	N/A	47% (BI-RADS)
Oliver et al. [42]	2005	Dense and fatty	Digitised SFM	MLO	320 (MIAS) and 300 (DDSM)	N/A	48% MIAS, 47% DDSM (BI-RADS)
Oliver et al. [43]	2007	Dense and fatty	Digitised SFM	MLO	322 (MIAS)	N/A	78% (BI-RADS)
Oliver et al. [44]	2008	Dense and fatty	Digitised SFM	MLO and CC	322 (MIAS) and 831 (DDSM)	Visually assessed	77% MIAS, 86% DDSM (BI-RADS)
Oliver et al. [45]	2006	Dense and fatty	Digitised SFM	MLO	322 (MIAS)	Visually assessed	82% (BI-RADS)
Torrent et al. [46]	2008	Dense and fatty	FFDM	MLO and CC	300	Visually assessed	82% MLO, 75% CC (BI-RADS)
Tortajada et al. [47]	2012	Dense and fatty	Digitised SFM and FFDM	MLO and CC	322 (MIAS), 831 (DDSM), and 236 (digital DB)	Visually assessed	86% MIAS, 77% DDSM, and 92% digital DB (BI-RADS)
Adaptive/modified fuzzy -means
Chen and Zwiggelaar [48]	2010	4 densities	Digitised SFM	MLO	N/A	Visually assessed	N/A
Keller et al. [49]	2011	2–9 densities	FFDM (processed)	MLO	160	Visually assessed; = 0.75 (automatic-manual)	N/A
Keller et al. [50]	2012	2–13 densities	FFDM (processed and raw)	MLO	160	Visually assessed; (raw, processed) = 0.82, 0.85 (automatic-manual)	N/A
Expectation-maximisation
Aylward et al. [51]	1998	Dense, fatty, and uncompressed-fatty	Digitised SFM	MLO and CC	70	Visually assessed	N/A
Zwiggelaar et al. [52]	2002	4–6 densities	Digitised SFM	MLO	263 (MIAS)	Visually assessed	67% (SCC)
Zwiggelaar et al. [53]	2003	4–6 densities	Digitised SFM	MLO	263 (MIAS)	Visually assessed	86% (SCC)
Zwiggelaar et al. [54]	2003	4–6 densities	Digitised SFM	MLO	263 (MIAS)	N/A	86% (BI-RADS)
Zwiggelaar and Denton [55]	2004	4 densities	Digitised SFM	MLO	36 (MIAS)	Visually assessed	75% (4 densities: 0%–10%, 11%–25%, 26%–50%, and 51%–75%)
Selvan et al. [56]	2006	5/8 densities	Digitised SFM	MLO	112 (mini-MIAS)	Visually assessed (92% good/excellent)	N/A