Background

There is an increasing level of interest in morphological and morphometric analyses, both for combination with molecular or ecological data and for a more thorough understanding of forms, e.g. investigations of shape spaces or functional morphology as well as phylogeny reconstruction (e.g. [1–7]). This is supported by modern data acquisition methodologies, mainly high resolution CT scans, which provide a multitude of characters on outer and inner surfaces. However, the approaches to landmark assignments have to be adjusted to the special situation of 3D data. 2D images of specimens allow for intersections of a structure, e.g. a suture, and the background of the image, while 3D objects have more degrees of freedom in rotation, so the same points would need a description such as e.g. being the anterior most point of a suture. A good software tool should therefore have additional options for navigating in the 3D space, but this is not fully provided by any of the currently available software packages (see software evaluation in Additional file 1). Here we describe such software which enables the user to load a stack of DICOM files, the standard output of medical imaging techniques, to calculate a volume rendering and to use a number of convenient tools for finding the optimal position for each landmark.

Results

Surface rendering versus volume rendering

Most current programs for digitization of 3D landmarks (e.g. [8]) rely on surface rendering rather than volume rendering. Surface rendering requires the definition of a threshold, i.e. a specification of the intensity of the surface of interest, and then renders a smooth surface passing through all voxels with the defined intensity. This process is based on the simplified model in which a single intensity can be used to specify all points on a specific surface. When this simplified model fails, for example because of varying properties of the surface (e.g. bone density), errors will be introduced into the rendering. These take typically the form of high noise when the threshold is too low, or non-physical holes through the rendered surface (pseudo-fenestrations) where the threshold is locally too high (Figure 1, left).

Figure 1
figure 1

Differences of surface rendering and volume rendering. Two different representations of the same skull rendered with two methods. Surface rendering (left) from Amira 5.2., volume rendering (right) from the TINA Manual Landmarking Tool. Note the pseudofenestration e.g. at the end of the incisor in the mandible in the surface rendering (arrows).

Full size image

In contrast, volume rendering operates by assigning each voxel an opacity based on its intensity, and a reflectivity based on an intensity gradient, and then modelling the passage of multiple light rays through the volume. In this way, every voxel is taken into account, mitigating problems with pseudo-fenestrations and noise, and providing the user with a more informative representation of the data (Figure 1, right). Although the problems of surface rendering are well understood, it is often preferred, because it requires less computing power. Volume rendering has traditionally been too slow to allow real-time user interaction; however, advances in accelerated 3D graphics hardware and in rendering techniques, such as the shear-warp algorithm [9] implemented in the Volpack library [10] used by the TINA Manual Landmark tool, have largely removed this limitation.

The working of the software

The TINA Manual Landmarking Tool (described in detail in Additional files 2 and 3) imports stacks of DICOM files or any subset thereof. The loaded data can be down-sampled by averaging across neighbouring voxels. For high-resolution datasets, down-sampling increases both the speed of the volume rendering and, through noise reduction, the quality of the rendered images. The scanned specimen can be displayed on four different windows, so-called Tv's, one giving a 3D visualization and the other three orthogonal 2D cross-sections (Figure 2, Additional file 4). The 3D Tv has several options for the volume rendering itself as well as for additional settings. For example, it is possible to render any specific surface, or to produce a translucent rendering of all surfaces simultaneously, to change the direction, intensity, and colour of the lighting, set the background colour, and use several other criteria for the display of specimens.

Figure 2
figure 2

User interface of the TINA Manual Landmarking Tool. Screenshot of the TINA Manual Landmarking Tool during digitizing session. See text for more details.

Full size image

The simultaneous display of the three 2D Tv's has an additional advantage in the use for multilayered structures, e.g. fish skulls, where the inner structures are covered by the outer ones, or noisy data, e.g. from specimens fixed in unbuffered formalin (Figure 3).

Figure 3
figure 3

Examples of multilayered objects. Screenshots of the 3D and 2D Tv tools. Left: Skull of a small fish, anterior tip of otolith selected. Right: Decalcified skull with precipitations in surrounding tissue of a European Ground Squirrel, first tip of molar teeth selected.

Full size image

If previous landmarks are available, they can be imported from a text file with running numbers and descriptions, both being visible during the digitization process (Additional file 5). Landmarks and their numbers can also be displayed in the 3D Tv. For bilaterally symmetric structures it is possible to load another file containing information about points on the sagittal plane and paired points on both sides, thus enabling the drawing of linking lines between corresponding landmarks. The latter two options also allow a quick, visual check that the points have been marked in correct order.

All four Tv's have a synchronized cursor, indicating the same position in all of them, which enables the user to very accurately pick extreme points, like the anterior most or dorsal most position of any structure, by moving the cursor into the desired direction in one window and then marking the first pixels of the structure to appear or disappear in another Tv. The software also supports the definition of an arbitrary plane and axis to increase the precision with which extremal points are identified (Additional file 6). A plane is defined by three landmarks, e.g. in the sagittal plane of a bilaterally symmetric structure, an axis by two end points that may or may not also be used as plane points. When both plane and axis are defined, the object can be rotated accordingly, thus allowing views from precisely defined angles in all spatial directions, as well as a more reproducible digitization of extreme points relative to the defined geometry.

Landmarks identified using the 3D Tv of the TINA Manual Landmark Tool will automatically be placed on top of the first structure whose density is above a user-defined threshold. The default setting corresponds to that of mouse skull bone in a CT scan, but it can be changed to any desired value. So, if e.g. a skull is scanned in a live specimen, the landmark would pass through air and soft tissue and would be placed on the first bony structure it meets, if the threshold is set accordingly. Although this is a very convenient supplementary feature for placing such landmarks, the use of a threshold in this way could introduce errors or bias, since the location of the landmark would be dictated by pixel intensity, with no account taken of the image noise or local variations in surface intensity. Therefore, the location can then be refined using the 2D windows by moving the cursor with mouse or arrow keys.

After all landmarks are placed, the resulting coordinates can be saved and exported as TPS or NTSYS files, which allow all further statistical analyses such as principal component analyses or canonical variates analyses using e.g. the IMP series [11], MorphoJ [12], or R [13].

The majority of functions can either be conducted with the mouse buttons or with keyboard shortcuts. Default options for both are provided, but they can be customized to a large extent, e.g. for left-handed users.

Discussion

The TINA Manual Landmarking Tool was developed in collaboration between computational scientists and a morphologist. This ensures on the one hand that computing power is optimally used and thus allows the implementation of the computational demanding, but for practical purposes superior volume rendering procedure on normal PCs. It ensures also that the tools that are offered reflect the needs of a practicing morphologist. To assess the performance of the software, we have compared the time and precision with which landmarks can be placed in positions corresponding to different types of landmarks ([14], Additional file 7), using either volume rendering plus the orthogonal cross-sections and defined axes in the TINA Manual Landmarking Tool (Additional file 8) or a typical surface rendering without defined axes in the commercially available software Amira 5.2 [8]. We found that a new user needs approximately the same time for locating landmarks in both software tools, but the precision of repeatedly identifying the same landmark is higher for landmarks placed at extremal points (type 3 landmarks according to [14]) with the TINA Manual Landmarking Tool (Additional files 9, 10).

Availability

A Knoppix CD version for testing purposes (i.e. to be run directly from the CD - no need for installation) can be downloaded from http://www.tina-vision.net/tina-knoppix/iso/ with a demo version including a mouse skull dataset with all appropriate settings, giving an impression of how the program works. This version supports also loading of datasets e.g. from a USB stick, although necessarily with relatively long loading times. Nonetheless, this enables a potential user to check whether the intended datasets work with the tool before performing a full installation of the software. Further information on new developments on this software is continuously updated at http://www.tina-vision.net.