A Review on Automatic Mammographic Density and Parenchymal Segmentation

Table 5

Summary of representative studies using statistical model building methods for mammographic tissue segmentation. Note that () this list consists of studies mainly related to Tabár tissue modelling and () in case of multiple results, only the best reported results are listed.

Study	Year	Number of mammographic building blocks	Modalities	Number of views	Number of images	Segmentation evaluation	Risk/density estimation accuracy
Texture statistical variation
Muhimmah et al. [71]	2007	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed	N/A
He et al. [72]	2008	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed	N/A
He et al. [73]	2009	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed (65% good/very good) Tabár (38% good/very good) BI-RADS	53% (Tabár), 70% (BI-RADS)
He et al. [74]	2011	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed	53% (Tabár)
He et al. [75]	2012	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed	85% (Tabár), 78% (BI-RADS)
He et al. [76]	2014	Nodular, homogeneous, and radiolucent	FFDM	MLO and CC	360	Visually assessed	79% (Tabár), 80% (BI-RADS)
Texture descriptors
He et al. [77]	2010	Linear, nodular, homogeneous, and radiolucent	Digitised SFM	MLO	320 (MIAS)	Visually assessed	78% (Tabár), 75% (BI-RADS)
He and Zwiggelaar [78]	2013	Nodular, homogeneous, and radiolucent	FFDM	MLO and CC	360	Visually assessed	N/A