Skip to main content

Measuring the knee

The knee is one of the largest and most important joints which, if damaged, can be replaced by an implant. Alex Ringenbach from the HLS has written software that can create an individual 3D model of femur and tibia bones from magnetic resonance images. This algorithm speeds up the making of precise patient-specific templates that enable surgeons to position and fit new knee implants more easily and accurately.

Fitting knee implants has been standard surgery for years. However even today, it still requires extensive planning due to anatomical differences between patients. An implant is an artificial object that must fit into a living system, the body. No two knees are the same and thus implants have to be positioned differently in each operation. In order to fit them accurately and ensure good joint function, surgeons today often use cutting templates. These attach to the femur and tibia and guide the surgeon when cutting through the bone to which the implant is to be fixed. To make these templates, a precise virtual 3D model is needed; Alex Ringenbach from the Institute of Medical and Analytical Technology creates this from Magnetic Resonance Images (MRI). For him, the advantage of this new method is clear: “In the past these templates were done by hand, which takes about five hours. We have written an algorithm that does the same job in less than ten seconds.” On this project Ringenbach is working with the firm Medivation AG in Brug, which integrates the segmentation algorithm into a planning tool for the production of templates.

There are several steps before the program generates the 3D blueprint for the cutting template. The main job of the algorithm is segmentation: the software must decide whether a point in an MRI image is bone, cartilage or other body tissue — a classic problem of machine image recognition, as Ringenbach explains: “It is much easier for a human being than a computer because we have prior knowledge. We know what the shape of the bone looks like and can complete our image of it in our head. We therefore also know what is bone and what is other tissue. From the signal data alone, which the MRI shows as shades of gray, this cannot be determined.”

Consequently the research team uses the Active Shape Model. From section images that a surgeon has segmented manually the software learns the shape of the bone and what the area near the bone surface probably looks like. From ten to twenty sets of data the researcher creates a bone model on the computer. This includes average shape, shape variation, information on the area around the bone and our prior knowledge. With this a picture can be reliably segmented.

In order to segment an image data set the model with the prior knowledge is applied: image values are put into the model, compared with the original, and the position and the shape of the model are adapted accordingly. “You have to be creative to find the right model parameters and to perfect the segmentation process,” says Ringenbach. Another challenge is the MRI examination itself: “Each patient lies in a different position in the MRI machine, which means that the bones are in a different part of the image, so there is no standard positioning system.” For the model, mathematical methods, the individual MRI datasets and the measured “average bones” must be harmonised as closely as possible. This process, which enables comparison of different data sets, is called registration and is the most time-consuming part of creating the 3D model. Due to the complicated mathematics it can take up to several days but it only needs to be done once.

Since signals are not standardized, data from different MRI machines often vary greatly. Hence there are also wide differences in the maximum intensity of individual image points — as in photos with different contrast levels. The new algorithm is programmed so that it can correct these variations in individual sectional images. Ringenbach considers the effort invested to be justified: “Knee implants often involve elderly people whose bones are fragile and have signs of wear such as spurs. If the templates do not fit exactly, they slip.” Ringenbach’s partner, Medivation AG, has developed a planning instrument with the algorithms to produce cutting blocks for the first successful operations. Further applications for the software are in development.

Methodology

  • Manual segmentation of MRI reference data
  • Registration of segmented surface data
  • Formation of statistical shape models (for femur and tibia)
  • Analysis of tissue texture in MRI data
  • Model development by incorporating texture classifiers
  • Model development by incorporating registration algorithms
  • Performance optimization by incorporating range limits

Infrastructure

  • Computer, programming language C ++

Support

  • Förderstiftung Technopark Aargau, Research Fund Canton Aargau

Collaboration

  • Medivation AG Collaboration Medivation AG 

FHNW School of Life Sciences

FHNW University of Applied Sciences and Arts Northwestern Switzerland School of Life Sciences Hofackerstrasse 30 CH - 4132 Muttenz
More information about the location
Share this page: