For each patch, some statistical features are computed and embedded in a vector with its histological type and its coordinates in the stereological grid. At this stage of the study, all features are obtained from global measurements on patches computed on

*RGB* color components (reduced to 64 values) and from the two first components (

*H*,

*E*) of the color deconvolution specific to Hematoxylin and Eosin staining [

11]. For any given component

*X*, the computed features are:

*X*,

*X* reverse sorting, cumulative_

*X*, 20%-40%-60%-80% quantiles of cumulative_

*X*, mean_

*X*, median_

*X*, mode_

*X*, Skewness_

*X*, Kurtosis_

*X*, PearsonModeSkewness_

*X*, that is a total of 13 data. Three of them are themselves histograms with 64 values but will provide a single measure after computing the distance between two signatures.With the 5 components (

*R*,

*G*,

*B*,

*H*,

*E*) 65 measures will be taken into account for a patch but 1010 values will be stored in its signature. Considering the sparse numerical range of features, the symmetric Kullback-Leibler distance has been retained for its ability to easily manage such values, while remaining fast to implement. The distance between two vectors

*p*
_{
1
},

*p*
_{
2
} of length

*n* is then defined by:

## Comments

No Comments