01.12.2018  Research article  Ausgabe 1/2018 Open Access
The same modality medical image registration with large deformation and clinical application based on adaptive diffeomorphic multiresolution demons
 Zeitschrift:
 BMC Medical Imaging > Ausgabe 1/2018
Background
Methods
Additive demons
Diffeomorphic deformation strategy
Exponential
ϕ = exp(
v)


Input: Velocity field v.
Output: Diffeomorphic map
ϕ = exp(
v)
Choose N such that 2
^{−N}
v → 0
e.g., such that max ‖2
^{−N}
v‖ ≤ 0.5 pixels
Scale velocity field
ϕ ← 2
^{−N}
v
for N times
do
ϕ ←
ϕ ∘
ϕ
end for

Algorithm implementation

Step 1: initialize displacement field.

Step 2: calculate demons driving force u(p) and update velocity field v.

Step 3: regulate deformation field using Gauss filter.

Step 4: obtain exponential mapping of deformation field by diffeomorphic transform.

Step 5: calculate similarity measurement function E(t) using Eq. 4.

Step 6: judge convergence condition.
Algorithm evaluation
Results
Synthetic image registration
The same modality medical image registration
Experiment Method

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


our method

514.7965

0.9993

0.9952

diffeomorphic demons

583.0147

0.9992

0.9937

additive demons

640.9294

0.9987

0.9906

active demons

1307.5

0.9944

0.9824

Experiment Method

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


our method

2257.2

0.9928

0.9762

diffeomorphic demons

2578.6

0.9905

0.9619

additive demons

2377.0

0.9920

0.9635

active demons

2332.2

0.9923

0.9644

Medical image registration with large deformation
Rotational Distortional Strength

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


30%

531.6939

0.9991

0.9950

50%

1763.9

0.9898

0.9479

70%

2948.6

0.9717

0.9107

90%

5046.0

0.9169

0.8136

Extrusion Strength

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


10%

543.2

0.9991

0.9940

30%

545.1

0.9991

0.9944

50%

529.8

0.9991

0.9949

70%

559.4

0.9990

0.9946

Clinical application
Experiment Method

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


our method

1371.9

0.9685

0.7911

diffeomorphic demons

1427.4

0.9678

0.7847

additive demons

1409.3

0.9385

0.7088

active demons

1401.8

0.9377

0.7043

standard result

1002.1

0.9874

0.8210

Analysis of different driving forces
Experiment Method

MSE/(A.U.)

NCC/(A.U.)

Structural Similarity/(A.U.)


Thirion

514.79

0.9993

0.9948

GaussNewton

476.1

0.9993

0.9948

Symmetric Demons

465.0

0.9994

0.9952

Influence of parameters on registration result
Registration Accuracy

σ
_{ x}



2.0

1.5

1.2

1.0

0.8


MSE/(A.U.)

464.81

479.2040

485.8915

508.2908

522.31

NCC/(A.U.)

0.9993

0.9993

0.9992

0.9992

0.9991

Structural Similarity/(A.U.)

0.9953

0.9949

0.9949

0.9943

0.9939

Time Consuming/(s)

136.46

150.37

193.94

222.06

287.1
