A review of automatic mass detection and segmentation in mammographic images.
Citations
724 citations
Cites background from "A review of automatic mass detectio..."
...As noted by Oliver and colleagues (20), there is no public available database made with digital mammograms....
[...]
...Good results can have been obtained in databases with ‘‘easy’’ cases, whereas bad accuracies may have been achieved by using ‘‘difficult’’ databases (16,18,20)....
[...]
...These types of annotations are not considered sufficient for some studies, as the one done by Oliver et al (20), where all circumscribed and spiculated lesions had to be manually segmented....
[...]
471 citations
Cites methods from "A review of automatic mass detectio..."
...Evidence for clinical benefit Texture analyses have been used extensively in x-ray mammography (52)....
[...]
323 citations
Cites methods from "A review of automatic mass detectio..."
...(i) Region-based methods (such as region growing, split/merge using quad-tree decomposition) in which similarities are detected, and (ii) boundary-based methods (such as thresholding, gradient edge detection) in which discontinuities are detected and linked to form region boundaries (Oliver et al., 2010)....
[...]
...…methods (such as region growing, split/merge using quad-tree decomposition) in which similarities are detected, and (ii) boundary-based methods (such as thresholding, gradient edge detection) in which discontinuities are detected and linked to form region boundaries (Oliver et al., 2010)....
[...]
254 citations
Cites background or methods from "A review of automatic mass detectio..."
...ratio of the mass visualisation, combined with the lack of consistent patterns of shape, size, appearance and location of breast masses (Oliver et al. (2010); Tang et al....
[...]
...ratio of the mass visualisation, combined with the lack of consistent patterns of shape, size, appearance and location of breast masses (Oliver et al. (2010); Tang et al. (2009))....
[...]
...Moreover, recently proposed segmentation methods (Rahmati et al. (2012); Cardoso et al....
[...]
253 citations
Cites methods from "A review of automatic mass detectio..."
...Region growing has been widely used in mammograms in order to extract the potential lesion from its background [49]....
[...]
References
24,320 citations
"A review of automatic mass detectio..." refers methods in this paper
...The traditional partitional clustering algorithm is the k-Means algorithm (MacQueen, 1967), which is characterised by simple implementation and low complexity....
[...]
22,840 citations
18,802 citations
"A review of automatic mass detectio..." refers methods in this paper
...Velthuizen used it to group pixels with similar grey-level values in the original images, while Chen and Lee used it over the set of local features extracted from the application of a multi-resolution wavelet transform and Markov Random Fields (MRF) analysis (Bishop, 2006). Moreover, the output of the FCM was the input of an Expectation Maximisation (EM) algorithm (Dempster et al., 1977) based on Gibbs Random Fields (Bishop, 2006). These final steps are closely related to the algorithm proposed by Comer et al. (1996). In contrast to FCM which improves k-Means using a fuzzy approach of the energy function, the Dogs and Rabbit (DaR) algorithm (McKenzie and Alder, 1994) performs a more robust seed placement. The DaR was used by Zheng et al. (1999b) and Zheng and Chan (2001) to obtain an initial set of regions which subsequently were used to initialise a MRF approach....
[...]
...Velthuizen used it to group pixels with similar grey-level values in the original images, while Chen and Lee used it over the set of local features extracted from the application of a multi-resolution wavelet transform and Markov Random Fields (MRF) analysis (Bishop, 2006). Moreover, the output of the FCM was the input of an Expectation Maximisation (EM) algorithm (Dempster et al., 1977) based on Gibbs Random Fields (Bishop, 2006). These final steps are closely related to the algorithm proposed by Comer et al. (1996). In contrast to FCM which improves k-Means using a fuzzy approach of the energy function, the Dogs and Rabbit (DaR) algorithm (McKenzie and Alder, 1994) performs a more robust seed placement. The DaR was used by Zheng et al. (1999b) and Zheng and Chan (2001) to obtain an initial set of regions which subsequently were used to initialise a MRF approach. As Li et al. stated (Li et al., 1995), MRF allow the modelling of joint distributions in terms of local spatial interactions, introducing thus, local region information into the algorithm. This information was also introduced in the work of Rogova et al. (1999) using a constrained stochastic relaxation algorithm with a disparity measure function, which estimated the similarity between two blocks of pixels in the feature space. In contrast, Cao et al. (2004a,b) used two information theory based clustering algorithms to segment masses. The first approach was the Deterministic Annealing approach (Rose, 1998), which is a global minimisation algorithm and incorporated ‘‘randomness” into the to be minimised energy function. In the second approach, they unified a fuzzy based clustering and Deterministic Annealing to obtain an improved algorithm. In contrast with these approaches, Bruynooghe (2006) segmented an enhanced image instead of the original mammogram. The enhanced image was obtained by removing the locally linear fine detail structure using a morphological algorithm based on successive geodesic openings (Davies, 1997) with linear structuring elements at various orientations. One of the earliest approaches to mass detection was the work of Brzakovic et al. (1990), which was based on a multi-resolution fuzzy pyramid linking approach, a data structure in which the input image formed the basis of the pyramid and each subsequent level (of lower resolution) was sequentially constructed. The links between each node and its four parents were propagated using a fuzzy function to upper levels. They demonstrated that this algorithm was directly correlated with the isodata clustering algorithm (Brzakovic et al., 1990). It has to be noted, that with this strategy, spatial information (region information) is taken into account. Like Fu and Mui (1981), we consider threshold methods as partitional clustering methods....
[...]
..., 1977) based on Gibbs Random Fields (Bishop, 2006)....
[...]
...Velthuizen used it to group pixels with similar grey-level values in the original images, while Chen and Lee used it over the set of local features extracted from the application of a multi-resolution wavelet transform and Markov Random Fields (MRF) analysis (Bishop, 2006)....
[...]
...Moreover, the output of the FCM was the input of an Expectation Maximisation (EM) algorithm (Dempster et al., 1977) based on Gibbs Random Fields (Bishop, 2006)....
[...]