A Vascular Feature Detection and Matching Method Based on Dual-Branch Fusion and Structure Enhancement

. 2024 Mar 15;24(6):1880. doi: 10.3390/s24061880

Algorithm 1

Input: Inner cavity image I, preprocessed inner cavity image I_pre, masked image I_mask
Output: Single-pixel vascular image I_ridge

1:
Branch 1 uses a blood vessel enhancement algorithm based on Frangi measure filtering on I_pre
2:
for each pixel p ϵ I_pre do
3:
for each scale σ do
4:
Calculate the Hessian matrix and its eigenvalues (λ₁, λ₂) and eigenvectors (I_x, I_y)
5:
Calculate Direction, R_B, and S based on eigenvalues and eigenvectors
6:
At scale σ Calculate the Frangi measure filtering result F(σ)
7:
end for
8:
Extract the maximum response and corresponding filtering result max(F(σ))
9:
end for
10:
Combining I_mask to generate Frangi measure vascular enhancement images I_Frangi, Direction
11:
Branch 2 using MFAT based vascular enhancement on I_pre
12:
for each pixel p ϵ I_pre do
13:
for each scale σ do
14:
Calculate the Hessian matrix and its eigenvalues (λ₁,λ₂) and Direction
15:
Set λ₃ = λ₂. Construct new eigenvalues based on the eigenvalues λ_p1, λ_p2, and calculate $F A T_{λ}^{σ}, R_{λ}^{σ}$
16:
At scale σ calculate and update MFAT results ${\bar{M F A T}}_{λ}$
17:
end for
18:
Remove pixels ${\bar{M F A T}}_{λ}$ < 1 × 10⁻²
19:
Extract the maximum response and corresponding scale filtering result max $({\bar{M F A T}}_{λ}$ (σ))
20:
end for
21:
for every steps μϵ (2, 4) do
22:
Establish a Gaussian model and calculate weights w₁ and w₂
23:
Weighted fusion of I_Frangi and I_MFAT to generate binary image I_vessel
24:
Using single-pixel algorithm based on I, I_vessel, and Direction
25:
Gaussian Filter Smoothing Image I
26:
Generate grid coordinate matrix [X, Y]
27:
for each coordinate (x, y) ϵ [X, Y] do
28:
Calculate the feature direction gradient ∇ and inner product
29:
If there is a sign change in the gradient along the direction of maximum curvature, it is considered as a vascular ridge line mask_ridge
30:
Multiply the ridge mask with the vascular image to obtain the vascular centerline image I_ridege
31:
if the single-pixel detection flag for blood vessels is true
32:
Interpolate and shift vascular images to obtain I_Rshift1 and I_Rshift2
33:
Compare to obtain a single-pixel ridge mask_ridge2 and its application
34:
end if
35:
end for
36:
Generate a single-pixel vascular image I_ridge
37:
Calculate the SSIM score between I and I_ridge
38:
end for
39:
Select the corresponding optimal parameter μ based on the maximum SSIM score
40:
Calculate the optimal weights w₁ and w₂
41:
Weighted fusion of I_Frangi and I_MFAT to generate binary image I_vessel
42:
Generate a single-pixel vascular image I_Frangi using a single-pixel algorithm based on I, I_vessel, and Direction