Fig. 11.23 Fibre tracts in the heart. Since water diffuses more along the fibres we can use DTI to show the direction of these fibres

The directions can also be displayed as lines, and we can join these lines in by tracking the most natural connection. These are fibre tracking images. Figure 11.23 shows a less common cardiac fibre tracking. It should be stressed that all stages of diffusion imaging introduce errors. At the fibre tracking stage, these may end up causing totally spurious lines. Thus fibre tracking results should be interpreted very cautiously.

11.6.3 Summary

This section illustrated manipulation of higher order data. It showed that we need to take account of the model of the physics producing the images in order to extract maximal information. It also illustrated something typical of medical images and perhaps less often considered in standard optical imaging. Medical images can be quantitative in the sense that they sometimes have units. Thus, instead of being just a visualization, they are also measurements. This has the advantage that they can be validated against other data: for example, data from non-imaging sensors.

11.7 Applications

We have shown how 3D images of human anatomy can be produced and analyzed to create 3D models of the patient. However, we still need to answer the question of clinical usability of such models. Now, we will briefly describe some applications, which show how 3D imaging can be used to influence treatment.

11.7.1 Diagnosis and Morphometry

The shape modeling example from Sect. 11.5.2.2 showed how such models can be used for segmentation. The reduced dimensional space of the shape model also

makes it possible to describe shape in a succinct way. This shape space provides a useful coordinate system for analysis of shape or morphometry. If diseased groups can be seen to occupy certain regions within shape space, then statistical shape analysis can be used for diagnosis. There have been numerous studies, for example, looking at the shape of brain structures such as the hippocampus in schizophrenia patients using statistical shape models.

11.7.2 Simulation and Training

It has long been agreed that the existing method of training surgeons is inadequate. The standard phrase that sums up the traditional method of surgical training was “see one, do one, teach one”. The learning curve of new surgeons is often significant and the results for patients may be catastrophic. This was particularly noticed with the introduction of keyhole or laparoscopic surgery. Other options for laparoscopic surgery include box trainers, for example suturing rubber gloves, but these are limited in scope. If a sufficiently high-fidelity virtual simulator can be developed, this has several advantages. The movements of the surgeon are known and measured, so scores of dexterity and ability can be derived. A significant database of cases can be created, including rare but important difficulties. This gives the surgeons experience that would otherwise take years in the normal apprenticeship model.

Simulation is a wide research field in itself. A simulator must model not only the 3D graphical data to produce a convincing view, but also the physics and motion of soft tissue as it deforms under the influence of surgical tools. The surgeon should ideally receive haptic information—touch feedback that is similar to the real situation. From a software point of view, there is a good research resource in this field, available from the SOFA2 network.

11.7.3 Surgical Planning and Guidance

Until recently, the standard way of viewing 3D medical images was as a series of slices printed on X-ray film and displayed on a light box. This has traditionally been the way such images are displayed to surgeons and, despite the availability of high resolution screens, it is still the norm for radiological images to be displayed in 2D. In surgery, the relationship between the preoperative imaging model and the patient is established entirely in the mind of the surgeon. The question is: given the 3D nature of the data, can we provide a more useful presentation of the patient to the surgeon?

The idea of image-guided surgery is that the preoperative model should be aligned to the physical space of the patient on the operating table. This has long

2www.sofa-framework.org.