Skip to main content
. 2023 Jun 28;25(7):991. doi: 10.3390/e25070991
Algorithm 1 Algorithm summary of the DBFS_GMI CAD System.
  • Require: 

    Image

 (a) Preprocessing

  • 1:

    Input: I

  • 2:

    Apply Fill the gaps I(x,y).

  • 3:

    Apply create Bounding Box by the image that circle the lesion I(x,y).

  • 4:

    Crop according to the coordinates of the bounding box Im(x,y)

  • 5:

    Output: RIm(x,y)

  • 6:

    Input: RIm(x,y)

  • 7:

    Apply Equalizes the histogram of the image, this improves the contrast RIm(x,y)

  • 8:

    Apply median filter, size = 5 × 5 to improve the edges of the lesion RIm(x,y)

  • 9:

    for RIm(x,y) do

  • 10:

        if RIm(x,y)≥IL

  • 11:

            Assign 1 to IThL(x,y)

  • 12:

        else if RIm(x,y)≤IL then

  • 13:

            Assign 0 to IThL(x,y) then

  • 14:

        end if

  • 15:

    end for

  • 16:

    Apply median filter, size = 9×9 to IThL(x,y)

  • 17:

    Apply apply the Suzuki–Abe method to IThL(x,y)

  • 18:

    Apply the label of the Suzuki–Abe method to find the mask of the lesion IThL(x,y)

  • 19:

    Output: IMask

 (b) Feature Extraction

  • 20:

    Input: IMask

  • 21:

    Compute Area, Perimeter, Circularity, Diameter and Eccentricity from Equations (27)–(30)

  • 22:

    Output: S handcraft features

  • 23:

    Input: RIm(x,y)

  • 24:

    Apply HOG technique to RIm(x,y) from Equations (5)–(7)

  • 25:

    Apply LBP technique to RIm(x,y) from Equations (8)–(19)

  • 26:

    Concatenate for f′=(x1′,x2′,...,x′n) and g′=(y′1,y2′,...,x′m) an obtain Fused Texture features

  • 27:

    Apply PCA to HL from Equations (20)–(26)

  • 28:

    Output: T Fused Texture features

  • 29:

    Input: RIm(x,y)

  • 30:

    Load the weights W from selected DenseNET-201 architecture

  • 31:

    Apply the weights W to RIm(x,y)

  • 32:

    Obtain the D deep learning features from (avg_pool)

  • 33:

    Output: D deep learning features

  • 34:

    Input: S, T, D

  • 35:

    Apply S∪T∪D to the extracted features

  • 36:

    Output: F Full set of extracted features

 (c) Feature Fusion

  • 37:

    Input: F for Mammography or Ultrasound

  • 38:

    Apply Mutual Information from Equations (32)–(35)

  • 39:

    Output: MF selected features

  • 40:

    Input: F for Mammography or Ultrasound

  • 41:

    Apply Genetic algorithm

  • 42:

    Output: UF selected features

  • 43:

    Input: F for Mammography and Ultrasound

  • 44:

    Apply Genetic algorithm and Mutual Information

  • 45:

    Apply UF∪MF to the selected features

 (d) Classification

  • 46:

    Input: UF selected features and MF selected features

  • 47:

    Apply UF∪MF to the extracted features

  • 48:

    Apply Class Separation Using the three classifiers

  • 49:

    Output: Diagnostic of the image I.