1.

pdf

alba_schez

TSIM - Advanced Methods in Medical Signals and Images

(MUIB)
4º Grado en Ingeniería Biomédica

Escuela Técnica Superior de Ingenieros de Telecomunicación

Universidad Politécnica de Madrid

Reservados todos los derechos.

No se permite la explotación económica ni la transformación de esta obra. Queda permitida la impresión en su totalidad.
 1.3 Active contours

No se permite la explotación económica ni la transformación de esta obra. Queda permitida la impresión en su totalidad.
1. Definition
Active contour is one of the active models in segmentation techniques, which makes use of the
energy constraints and forces in the image for separation of region of interest. Active contour
defines a separate boundary or curvature for the regions of target object for segmentation. The
contour depends on various constraints based on which they are classified into different types
such as gradient vector flow, balloon and geometric models. Active contour models are used in
various image processing applications specifically in medical image processing. In medical
imaging, active contours are used in segmentation of regions from different medical images such
as brain CT images, MRI images of different organs, cardiac images and different images of
regions in the human body. Active contours can also be used in motion tracking and stereo

Reservados todos los derechos.

tracking. Thus, the active contour segmentation is used for the separation of pixels of interest
for different image processing.
An energy functional is associated with the contour, so, the problem of finding the
object is cast as an energy minimization problem. So, there are two different ways for
segmenting images:
• Classical approach: link edge segments together.
• Active contours: a smooth curve which matches to images data. These methods allow us to
do a correct segmentation of objects with diffuse boundaries. We have two approaches:
explicit contours or implicit contours.
the different between the classical approach is that in theclassical one, the segments need to be "together" to get a valid process.
With active contours there is an evolutio.

2. Explicit representation
The contour is defined in the image domain as a parametric curve.
Explicit geometry: parametric curve
Implicit geometry: level set
2.1 Snakes
In medical image analysis, the term “snake” is usually associated with parametric
representation of contours. A snake is an energy minimizing, deformable parametric curve
guided by constraint and image forces that pull it towards object features (i.e., contours). In
other words, a snake is an energy-minimizing spline guided by external constraint forces and
influences by image forces that pull it toward features such as lines and edges. Snakes are active
contour models, and they provide a unified account of a number of visual problems, including
detection of edges, lines, and subjective contours. It is typically used in dynamics images and
couple active contours.
The energies that we can find over the snakes are:
• Internals: Constrains the shape of the snake (e.g., smoothness). It is minimal when the
shape is similar to the shape of the sought object.
• Externals: Search for suitable image features (e.g., edges, gray level, texture…). Minimal
when the snake is at the object boundary position.
• Restriction energies: external constraints useful for particular cases: for attraction force
towards certain point or region and for user interaction. A priori knowledge about object,
allowing to extract boundary even if some image data is missing.

a64b0469ff35958ef4ab887a898bd50bdfbbe91a-3713972
 A) INTERNAL ENERGIES
The splice energy is composed of a first-order term (elastic energy) and a second-order
term (curvature energy). In the first-order derivative, the snake is treated as an elastic rubber
band possessing elastic potential energy, and it discourages stretching by introducing tension.
Whereas in the second order derivative, the snakes also behave like a thin metal strip were
any deformation is restricted (stiffness/Bending), so it allows to avoid corners and smoothness.
By adjusting the weights alfa and beta, the relative importance of the band and the thin strip is
controlled. Setting beta zero at a point allows the snake to become second-order

No se permite la explotación económica ni la transformación de esta obra. Queda permitida la impresión en su totalidad.
discontinuous and develop a corner.

B) EXTERNAL ENERGY
Describes how well the curve matches the image data locally. It is driven by the gray
level, in a way that the snake will be attracted to the lightest or darkest nearby contour. When
it finds edges, the snake tends to converge on edges (following gradients pattern), so it gets a
negative sign, minimizing energy forces.
if we are getting a large gradient, the energy willbe small, since the obkective is to get the minimum value of the total energy

C) MINIMIZATION
This process consists of finding the curve that minimize the energy functional. A
minimum of the energy functional implies that its derivative is equal to zero.

Reservados todos los derechos.

The algorithm is based on a gradient descent algorithm, in which the derivative of
energy with respect to distance (force) is computed, which allows efficient optimization
converging to a local minimum. At each iteration we compute a new snake position within
proximity to the previous one. New snake energy should be smaller, and it stops when the
energy cannot be decreased within local neighbourhood (local minima).
local minima can be the right segmentation or not

D) DISCRETE MODEL: we arent working in the continuous domain any more

The snake is represented by N control points, which are known as snaxles, so the snake
discretized in the control points (snaxles).

E) REMARKS
The benefits that present are:
• Low algorithmic complexity (e.g., compared with level sets)
• Can account for open as well as closed curves
• Can Support highly intuitive interaction mechanisms
• Easy to introduce prior knowledge about objects

On the other hand, the drawbacks are:

• Topology of the object of interest should be known
• Self-intersections and topological changes
• Hard to extend to 3D and 4D data
• Sensible to initial conditions
• Very sensitive to false local minima which leads to wrong convergence

si lees esto me debes un besito

a64b0469ff35958ef4ab887a898bd50bdfbbe91a-3713972
 Drawbacks of classic snakes:

- Curve defined by a set of disconnected points

- Slow convergence due to large number of points
- High order derivative on the discrete curve may not be accurate in noisy
environments
this leads to interpolation curves and approcimation curves
(VER LAS DIAPOS QUE FALTAN DE ESTO; a partir de la 22 del ppt de contour)

No se permite la explotación económica ni la transformación de esta obra. Queda permitida la impresión en su totalidad.
3. Implicit representation: level set
A limitation of active contour based on parametric curves of the form (using snakes) is that it is
challenging to change the topology of the curve as it evolves. If the shape changes dramatically,
curve reparameterization may also be required. So, al alternative representation for such closed
contour is to use level sets (LS).
LS evolve to fit and track objects of interest by modifying the underlying embedding function
instead of curve function. It is an efficient algorithm, in sense that a small strip around the
locations of the current zero-crossing needs to update at each step. It is based on topological
changes.
A curve can be implicitly defined as a level set of a surface of higher dimension. The contour
(level set 0) is the intersection of a surface of higher dimension (level set function) with an image.

Reservados todos los derechos.

The level set will have a specific evolution, in which the central idea is to track a propagating
interface by embedding it as the zero-level set of a higher-dimensional function.

3.1 Level sets minimization: iterative algorithm

From an initial curve, in this algorithm, it is necessary to construct a surface that contains the
curve as level zero (Usually, signed distance function is used). After that, it will perform surface
evolution. Finally, the last step is to adjust the surface (calculate signed distance again).
There are numerous methods to speed up level set methods, because sometimes we need to
play with volumetric data of 512x512x64 voxels.

3.2 Level sets: advances techniques

In these techniques, a priori information about the regions shape is known, so we are working
with knowledge-based segmentation. To obtain the priori information, it is necessary to do a
training test, in order words, segmentation and registration of training sets, in order to obtain a
model that express the set variability. Once we get it, this model should be included in the
distance function. For example, in myocardial segmentation:

3.3 Level sets: remarks

Advantages of levels sets:
• Discrete implementation in a lattice (digital images)
• Allow topological changes in the curve evolution
• Can Make use of a priori knowledge of object shape
• Easily extended to n-dimensional data
Drawbacks:
• Closed curves

si lees esto me debes un besito

a64b0469ff35958ef4ab887a898bd50bdfbbe91a-3713972
a64b0469ff35958ef4ab887a898bd50bdfbbe91a-3713972
•
•
Computationally expensive
Hard to find stop criteria in minimization algorithm

4
Reservados todos los derechos.
No se permite la explotación económica ni la transformación de esta obra. Queda permitida la impresión en su totalidad.