Image segmentation algorithm and data analysis

In these section we propose an algorithm for the characterization of carcinogenic microcalcifications by morphology descriptor BI-RAD 4. the algorithm is implemented by means of a two-phase algorithm; The first phase is the segmentation of images where suspicious microcalcifications are obtained, and the second phase these objects are classified for diagnosis using morphology descriptors BI-RAD 4.

Phase 1: segmentation

An improved non-linear based filter using enhanced factor for image thresholding algorithm

The proposal:

Let f(x, y) the source image, and g(x, y) = α output image, where α is the enhanced factor defined by:

α ← t / (m – t)               (1)
α ← m – (v * (1+ α))    (2)
α ← α – ( α * t)            (3)


in (1), t is the threshold value that minimizes the weighted within-class variance of f(x,y), m = max( f( x, y ) ) and v = med { f ( xi, yj ), i, j ϵ W }, where W is n X n sub-matrix whit n = 3, Noise, brightness and contrast level are automatic adjusted by (2) and (3) respectively, then each object is labeled.

Phase 2: analysis

A propose algorithm for the characterization of carcinogenic microcalcifications by distribution and morphology descriptor BI-RAD 4.

The proposal:

Let be A a k X m the matrix of morphological variable values for k suspicious object in g(x,y), B a n X m the proposed matrix, m the set of variables and n = 2m, posibles values depending:

where li, lc, ls are minimal, central and maximun posible variable value. Additionally each row of B is previously labeled with 0 benign 1 malignant in the L set. The k-esime object in A, is classified Lk, if Akm = = Bkm

Associated files

USTDS - Technical support and software development unit