Table 4 Feature Extortion

1

can.u

Co-ordinates of a nodule candidate(horizontal)

2

can.v

Co-ordinates of a nodule candidate(vertical)

3

can.Grad₁

Co-ordinates of a nodule likelihood values

4

can.CV₁

Computed by using gray level values

5

can.Grad₂

Co-ordinates of a nodule likelihood values

6

can.CV₂

Computed by using gray level values

7

Shape₁

Area for a segmented nodule candidate(A_region)

8

Shape₂

Short and long axes of an ellipse which are robust to nodule candidate

9

Shape₃

A_region/A_{convex hull}

10

Shape₄

[d_{candidate- center} /squarero ot(Shape₁/Л)]

11

Gray₁

μ^region – μ^surround

The gray-level feature was estimated using fragmented candidates and their surrounding regions in both pre-processed image and nodule-enhanced image. A surrounding region was constructed by subtracting a candidate region from a dilated candidate region. μ^region➔ Mean of a region

μ^surround➔Mean

of a surrounding region

12

Gray₂

σ^region- σ^surround

σ^region➔Standard deviation of a region

σ^surround➔Standard deviation of a surrounding region

13

Gray₃

min^region-min^surround

min^region➔ Minimum value of a region

min^surround➔Minimum value of a surrounding region

14

Gray₄

max^region-max^surround

max^region ➔ Maximum value of a region

max^surround➔Maximum value of a surrounding region

15

Gray₅

Calculated using Gray₁

16

Gray₆

Calculated using Gray₂

17

Gray₇

Calculated using Gray₃

18

Gray₈

Calculated using Gray₄

19

Grad₁

\( \overline{\mathrm{Gr}}=\left(1/8\right)\sum \limits_{\mathrm{k}=0}^7{\mathrm{Gr}}^{\mathrm{h}} \)

\( {\displaystyle \begin{array}{c}{\mathrm{Gr}}^{\mathrm{h}}=\left[1/{\mathrm{N}}_{\mathrm{h}}\right]\sum \cos \kern0.28em {\alpha}_{\mathrm{mn}}\\ {}\mathrm{mn}\in {\mathrm{region}}_{\mathrm{h}}\\ {}{\mathrm{t}}_1\le {\mathrm{M}}_{\mathrm{mn}}\le {\mathrm{t}}_2\end{array}} \)

where N_h is number of pixels in segmented candidate area h and cos α_mn denotes likelihood values used in two stage nodule enhancement method

20

Grad₂

\( \upsigma =\surd \sum \limits_{\mathrm{k}=0}^7{\left({\mathrm{Gr}}^{\mathrm{h}}-\mathrm{Gr}\right)}^2 \)

21

Grad₃

\( =\overline{\mathrm{Gr}}/\sigma \)

\( \overline{\mathrm{Gr}} \) denotes gradient feature

22

Surface₁

λ_min

Segmented candidate area in nodule enhanced image was robust to fourth order bivariate polynomial. The principal curvatures was computed at highest elevation point in the candidate region.

23

Surface₂

λ_max

24

Surface₃

λ_min λ_max

25

Texture₁

∑ [C(i,j)²]

ij

C(i,j) ➔Co-occurrence matrix computed over neighboring pixel and a summation range from minimum to maximum pixel value in pre-processed image

26

Texture₂

∑ (i-j)²C(i,j)²

ij

C(i,j) ➔Co-occurrence matrix calculated over neighboring pixels and a summation range from the minimum to the maximum pixel value in the pre-processed image.

27

Texture₃

Calculated based on Texture₁ and Texture₂

28

Texture₄

Calculated based on Texture₁ and Texture₂

29

Texture₅

Calculated based on Texture₁ and Texture₂

30

Texture₆

Calculated based on Texture₁ and Texture₂

31

False Positive

Loverlap / Lregion Where Lregion is length of boundary of a candidate area and Loverlap is number of pixels on boundary that overlap edge chain.