3.1. Segmentation and visualization

Since decades, the fully automatic segmentation of images—not only in the medical domain is a challenge that is still unsolved in general [23, 35]. Not only the algorithms are the topic of research but also the question, how to evaluate the result of an image segmentation [33, 34]. Many modern segmentation algorithms in medical image processing use level sets; a review of these techniques can be found in [32]. State-of-the-art systems use the input of an expert to initialize segmentation and also allow for manual correction of the segmentation results.

To provide the input for modeling, the shape of the bones of a patient is required in three dimensions, which requires accurate segmentation. The key to success is to provide a computer assistance to speed up segmentation and to provide an interface that allows for fast and simple computer tools.

Bone structures are segmented in CT data. In those cases, where an MR image is also available, CT data and MR data are registered [36, 37] and fused. Most modern medical workstations provide such tools.

#### 3.2. Defining biomechanical parameters

head, should be avoided. This aim can be achieved by optimized positioning of the implant, an

The goal of finding an optimal positioning (1) of an implant is in focus, which ensures, for example, a "natural" stress distribution, can be determined by the patient-specific simulation model. If desired, the exact coordinates of the positioned implant can be exported. These can then be taken into account in the operation planning. The rapid prototyping can be realized

If the baseplate design of a "standardized" implant is not suited to the morphological conditions of the contact surface of a patient-specific vertebral body, alternative baseplate designs can be demonstrated by means of a biomechanical simulation. In addition, corresponding modifications in implant material properties can also be analyzed and its effect on the spinal structures can be evaluated (see Figure 1 process chain, loop (2a)). Thus, the risk of complications is minimized through, for example, an insufficient anchoring and the concomitant loosening of the implant or the occurrence of load peaks by point contact. On the basis of the simulation results, the implant can be re-designed specifically for a patient (see Figure 1 process chain, loop (2b)), and corresponding input data (see Figure 1 process chain, loop (2c))

In the field of rapid prototyping, generative creations and processes with material removal can be different. Generative creation is the production of 3D physical models by applying material in thin layers and solidifying them. Established techniques are stereolithography, selective laser sintering, fused deposition modeling, laminated object manufacturing, and inkjet-print-

With such an expanded process chain, patient-specific implants can be produced not only with optimally shaped contact surfaces, which ensure a permanent fit of the implant without sinking and slipping, but also preoperative predictions can be made about the biomechanical

Medical imaging is used as a basis for biomechanical modeling of the spine. Depending on the scientific question, these may be obtained from MRI cone beam computer tomography (CBCT), positron emission tomography (PET), single photon emission computed tomography (SPECT), and ultrasonography (US) [3]. From appropriate image data, using a specific image post-processing algorithm, 3D visualizations can be generated as well as an analysis of the

Since decades, the fully automatic segmentation of images—not only in the medical domain is a challenge that is still unsolved in general [23, 35]. Not only the algorithms are the topic of research but also the question, how to evaluate the result of an image segmentation [33, 34].

optimization of the implant design as well as of the implant material properties.

can be generated for the further processing of rapid prototyping.

ing techniques. A more detailed explanation can be taken from [3].

effects of the implant and an optimized positioning can be proposed.

sequentially or in parallel.

50 Innovations in Spinal Deformities and Postural Disorders

3. Medical imaging

biomechanical behavior of the tissue.

3.1. Segmentation and visualization

One possible way to determine the mechanical behavior of the spinal bony structures is to determine them from CT data. The radiation emitted during the scanning process penetrates the object and is weakened to varying degrees by the tissue. On the basis of the measured intensity reduction, an attenuation coefficient is calculated for each beam direction, to which a CT number is assigned. The unit of the CT number is given in Hounsfield [Hu] [6]. The Hounsfield unit is defined by the following equation:

$$H = 1000 \ast \frac{\mathcal{C}T - \mathcal{C}T\_w}{\mathcal{C}T\_w - \mathcal{C}T\_a} \tag{1}$$

where CTw and CTa are the CT values of water and air [7]. According to Sun et al. [8], the CT number can be correlated to the density, for example, of the bone by a linear interpolation using relations available in published literature. The determined density can then be related to the so-called Young's modulus E. The heterogeneous elasticity, for example, of the cancellous and cortical bone, can in turn be defined by the following relationships [5]:

For cancellous bone, CT < 816:

$$
\rho = 1.9 \ast 10E - 3 \ast CT + 0.105 \tag{2}
$$

$$E = 0.06 + 0.9 \ast \rho \ast 2\tag{3}$$

For cortical bone CT > 816:

$$
\rho = 7.69 \ast 10 - E \ast 4 \ast \text{CT} + 1.028 \tag{4}
$$

$$E = 0.09 + 0.9\rho 7.4\tag{5}$$

In addition to determining biomechanical parameters by means of the CT number, it is also possible to obtain biomechanical information of, for example, the degree of intervertebral disc degeneration by a radiographic grading system. To determine more objective assessment of lumbar and cervical intervertebral disc degeneration, a new radiographic grading system [9, 10] is developed. The classification of this radiographic grading system is based on the three variables "height loss," "steophyte formation," and "diffuse sclerosis." According to Wilke et al. [9], each of these three variables first has to be graded individually on lateral and posteroanterior radiographs. Finally, the so-called overall degree of degeneration is assigned on a fourpoint scale from 0 (no degeneration) to 3 (severe degeneration).

Recent approaches try to estimate biomechanical properties of humans by tracking motions both in color image sequences as in distance measurements. Such data are available from devices that were designed for consumer games, but have also been used experimentally in medical applications, for example, in Ref. [38].
