Image Segmentation

In image analysis and medical diagnosis, segmentation techniques are used to separate the objects of interest from the surrounding context. Once a segmentation has been computed, several statistics can easily be derived from the object.

In general, manual segmentation methods are very time-consuming. Therefore, our focus is on semi-automatic and fully automatic methods. We concentrate on the minimization of user interaction in order to keep things as simple as possible while providing the results as fast as possible. To achieve this goal, we take advantage of latest developments in computer graphics hardware for noticeable performance speedups.

Features

The major challenges of medical image segmentation arise from the high variability within the data. Simple segmentation techniques can only be used if the structure is very homogeneous. Furthermore, most of the problems encountered in medical imaging are very specific. Generic automatic algorithms are often not suitable to achieve optimal results. We therefore provide fully automatic algorithms that can be customized to fit the needs of a specific application field, as well as very generally applicable semi-automatic algorithms that rely on a minimal user input to find the desired regions.

The following list summarizes the main features of our segmentation techniques.

Automatic Segmentation

  • Multi-modal input data
  • Single-click solutions
  • No parameterizations necessary
  • Tumor segmentation in PET and SPECT
  • Bone splinter segmentation

Semi-Automatic Segmentation

  • Support for all kinds of input data
  • Heterogeneous objects
  • Manual refinement of segmentation result
  • Sparse set of seed points for full 3-D segmentation
  • GPU hardware acceleration

Examples

Bone Segmentation in MR Images

The left MR image shows the segmentation of a human femur. The heterogenous bone tissue in the MR image has been segmented using our semi-automatic segmentation method that requires only few marker points inside and outside the structure. The computation of the 3-D segmentation result takes only a few seconds on current GPU hardware.

The middle image shows a volume rendering of the MR scan with the 3-D surface of the femur segmentation.

The image on the right shows the surfaces of the segmented patella and tibia in addition. The segmentation results can then be used for further processing.

 

Kidney Segmentation in CT Images

The left image shows the segmentation of a kidney in a CT volume data set. Here, our semi-automatic algorithm was used on heterogeneous soft tissue.

The middle image shows the volume rendering of the CT image with the computed kidney surface.

The video in the right column demonstrates the segmentation procedure. Note that the playback time is the same as the recording time. The user has to select only a few seed points inside and outside the organ. The seed points are also only defined in one single slice of the kidney, which is enough to create a fully 3-D segmentation result. The colored overlay shows the generated 3D mask of the segmentation result.

The anatomical structures of the renal pyramid, the renal pelvis and renal column of the kidney constitute an image region with quite heterogeneous intensities. The semi-automatic segmentation algorithm is able deal with such regions of interest and allows further refinements by the user, in case a first segementation result was not fully satisfactory.

 

Ventricle Volume Measurement

The segmentation of ventricles can be used to compute the blood volume of the heart. A suitable measure to determine the heart's efficiency is the ejection fraction, which is the amount of pumped blood divided by the amount of blood in the ventricles. From this measure, the physician can make estimations about the function of the heart.

The left and middle CT images show a segmentation result for a ventricle. The segmented ventricle is bounded at the cardiac valve where the next segmented region starts. The red contour shows the segmented aorta.

The right image shows the 3-D visualization of the generated surface models of each segmentation using surface and clipped volume rendering.

3D Bone and Organ Model for 3D Printing

The following 3D STL model has been created with a semi-automatic smart brush.

Send an E-Mail to info@chimaera.de with subject "STL Model" to get a download link of the model.