Info

C├Ątatale Pad S pdcmg J

Cois

Slice Spacing jocoaoc (rt-,m)

Slice Spacing jocoaoc (rt-,m)

Figure 6.2 BioBuild can load raw image data, allowing the user to enter any missing volume properties

Data can also be loaded from a remote DICOM server, using the DICOM medical imaging network protocol. Remote server access is easily configured for any number of servers, as highlighted in Figure 6.3.

Many hospitals now use the DICOM standard for storage and transmission of patient data from multiple imaging modalities, and BioBuild is able to access these patient data seamlessly across the hospital network. Patient searches are made even easier by masks, such as the first letter of the patient's name, the patient's hospital ID or even the patient's doctor.

As DICOM is now the standard storage format for most patient data, BioBuild has been tested for compatibility against a wide variety of scanner types, as there are often vendor-specific peculiarities. Further, datasets of all resolutions and sizes are supported. Currently, a typical dataset consists of a voxel resolution of approximately 0.4 mm in both X and Y planes,

Figure 6.3 Patient data can be loaded directly from any DICOM server

and a slice spacing (Z plane resolution) of 0.3-1.5 mm. Traditionally, these were considered high-resolution values. However, with the latest model scanners, in particular multislice CT systems, such resolutions are now standard. Conversely, with higher-resolution scans comes the need for high-specification workstations to process the data in a reasonable amount of time.

The next step after loading a dataset is to inspect and edit the volume. However, before editing a volume it can be useful to reduce the overall volume size, as generally only a part of the scan will be used for biomodel production, and often scans contain a lot of unwanted information (air surrounding the head, for example). Reducing volume size can have a significant impact on processing speed, especially on low-end workstations. BioBuild has several powerful features focusing on these key aspects.

6.2.2 Volume reduction

After the desired patient study is selected, the user can preview the images via a thumbnail snapshot. The user then restricts the loaded volume size interactively by applying a clipping rectangle or freeform region that collimates the volume to include only the region of interest. This is a very useful feature for reducing overall image volume sizes and processing times, particularly when the region of interest is comparatively small to the total scan area.

6.2.3 Anatomical orientation confirmation

Once the desired scan region has been selected, and the volume reduced, the user is presented with two views of the data: the default slice data as stored in the file (usually in the axial plane) and a plane reformatted from the slice data at 90┬░ (usually the sagittal plane). The user then scrolls through each series of images to peruse the orientation of familiar anatomical structures. If the anatomy in the images does not match the orientation annotation, the images can be flipped top-to-bottom and/or left-to-right using arrow buttons located within each window. The user then confirms the orientation by clicking a button and the series is constructed into a new BioBuild volume.

Orientation confirmation is a critical step in the biomodelling process, as scans imported into BioBuild are removed from their native environment and format. Although BioBuild is designed with this in mind, it is also imperative that a trained radiographer or imaging technician assess and confirm the scan orientation to ensure that the produced biomodel will be anatomically correct. Owing to the serious implications of a mirrored or otherwise incorrect biomodel, this step is enforced on the user and must be completed before any volume editing operations can take place.

Once confident the scans are in the correct orientation, the next critical step in biomodel production is to inspect the volume.

6.2.4 Volume inspection and intensity thresholding

The volume dataset is displayed as a 2D transverse view, in a single pane. Typically, the user scrolls through the data, slice by slice, using the mouse or arrow keys to confirm that:

  • no patient movement has occurred during the scan;
  • all structures of interest are included in the scan area.

Figure 6.4 2D viewing window allows for multiple, simultaneous views of the volume. The large window on the left shows the transverse view, the upper right window shows a coronal view and the lower left window shows a sagittal view. The position of each window with respect to the others is highlighted by a green line. The separate threshold window is also visible in the top right corner, and two floating toolbars are visible

Figure 6.4 2D viewing window allows for multiple, simultaneous views of the volume. The large window on the left shows the transverse view, the upper right window shows a coronal view and the lower left window shows a sagittal view. The position of each window with respect to the others is highlighted by a green line. The separate threshold window is also visible in the top right corner, and two floating toolbars are visible

The user can repeat this process in the other two primary views (sagittal and coronal) to confirm the quality and suitability of the scan data for the intended purpose. This is important, as the quality of the physical biomodel is in large part determined by the quality of the original image volume data. The data can then also optionally be viewed in a multipaned window, displaying the three orthogonal views simultaneously - generally with transverse/axial as the main view, and coronal and sagittal secondary views. A window displaying all three views is shown in Figure 6.4.

6.2.4.1 Intensity thresholding

A crucial aspect of the inspection of greyscale image volumes for 3D imaging and biomodelling is intensity thresholding. The threshold determines the set of structures to be included in the rendering and/or biomodel. In the single or multipaned view, the intensity threshold level and window can be adjusted in real time, serving to highlight the desired structures for both 3D rendering and biomodelling. Further, contours can optionally be displayed, also updating in real time in response to threshold adjustments, which indicate regions of connectivity in the data. Contours are displayed as green lines bounding structures in the data. The contours provide an excellent aid to selecting an appropriate threshold; however, they may be disabled for a more traditional view of the image slices. The threshold is selected empirically by the user via inspection of the data and slice contours.

6.2.4.2 Display options

Options in the 2D view are available via a right-mouse click and include zooming, toggling segmentation contours on or off, toggling display smoothing on or off and opening the threshold dialogue.

Smoothing in this case refers to the blurring of the pixels in the data matrix, as illustrated in Figure 6.5. If smoothing is turned off, the individual pixels become clear. With smoothing turned on, discrete pixels are no longer discernable. The effect of smoothing is particularly noticeable when the slice data is viewed at a high zoom level.

6.2.5 Volume editing

Volume editing is the most powerful and perhaps most important feature of BioBuild, as it is the critical step in the production of a biomodel. It is this feature that allows a user to remove unwanted structures from a dataset, select the region of interest and tailor the biomodel as necessary.

Figure 6.6 shows a typical workspace set-up. All volume edits take place in the 2D slice window (left), and the final biomodel is inspected to ensure it contains all important structures (right).

The more commonly used volume editing options include the following:

  • creating a 'segmentation region' via voxel connectivity at a selected threshold;
  • adding and removing material;
  • coloured image overlays highlighting edit areas and 'segmentation regions';
  • creating a 'segmentation region' via the 'connect and keep' operation;
  • selecting 'seed point' in an area of interest with the brush tool;
  • selecting 'grow region' to initiate connectivity and automatically create a region containing all voxels connected to the seed point;
Figure 6.5 Smoothing illustration. The left image shows smoothing turned on, the right image shows smoothing turned off. The images are zoomed to 800% of normal
Figure 6.6 BioBuild multiwindowed edit and view interface
  • deleting all voxels not connected to the current voxel selection;
  • finding the smallest volume containing all selected voxels.

These, and more advanced volume editing features, are outlined below.

6.2.5.1 Connectivity options

Producing accurate and useful biomodels is closely linked to the ability to add and remove structures in the volume. BioBuild provides the powerful ability to limit or extend structures quickly and easily with its connectivity options. These include adding material to a volume to create bridges and support struts (see Figure 6.7), removing unnecessary structures and a combination of the two, which, for example, allows bone to be removed while leaving vessels intact.

6.2.5.2 Volume morphology

Volume morphology comprises the erosion and dilation of a volume, which is the expansion or contraction of a contour by a uniform mask. Consequently, the behaviour of the morphology operations is closely tied to the current threshold value. Further, open and close operations are defined. An open operation first performs an erosion followed by a dilation. A close operation

Figure 6.7 Artificial structures are easily added to biomodels. The case shown illustrates adding material to connect two bony structures. In the left window one of the artificial structures is clearly visible. In the right window the resulting structures can be easily inspected

performs a dilation followed by an erosion. Erosion and dilation have the effect of smoothing the contour lines.

Before performing morphological operations, the volume may first need to be resampled to a finer resolution. This is because BioBuild will default to the coarsest dimension to determine the mask size. If the mask is too large, the effect on the contours can be too dramatic, affecting the structure of the biomodel to an extent much greater than is usually intended.

6.2.5.3 Region morphology

Region morphology allows the user to selectively choose the structures to erode or dilate. That is, specific structures of different threshold, such as blood vessels or thin bone, can be selectively modified as desired.

6.2.5.4 Volume algebra

Often it is extremely useful to subtract a modified copy of one volume from the original. This operation, along with the union and intersection of volumes, can all be performed using the volume algebra operations of BioBuild.

For example, volume subtraction can be used to remove bone in CT angiography scans. This is easily achieved with the following steps. First, the volume is copied, creating two volumes of the same dataset. Then, the threshold is modified in the copied volume to select bone. A slightly low threshold is used to ensure no vessels are unintentionally removed. The bone is then selected by seeding a point on any bony structure with the paint brush tool and starting the 'region grow' operation. This causes all voxels containing bone, and connected to the seed point, to become selected. Because a low threshold was used, the bone is now dilated to compensate. The selection is then inverted and the 'remove material' operation is performed. By inverting the selection, everything except bone is removed. Finally, the copied and edited volume is subtracted from the original, leaving only the vessels. Figure 6.8 illustrates this process. The left image shows the original skull base including vessels. The right image shows the vessels with bone removed.

The dilation step is important as a low threshold was used initially to select the bony structures. This means that not all the bone may be included. Dilating ensures that, when the copied volume is subtracted from the original, no bone remains.

6.2.5.5 Labels

It is possible to label biomodels automatically, using the 'text' feature. By adding text to a volume, a separate STL file is created that represents the desired text. It can also be given a different intensity to differentiate it from the anatomical model. When the biomodel is built physically, the text can appear as a different colour on the biomodel, and more than suffices as a labelling technique. To allow labelling in non-colourable RP systems, the text can be punched into the model using the 'remove material' editing option, or added as a raised label using the 'add material' editing option.

6.2.5.6 Volume transformations

If a scaled or otherwise transformed model is required, BioBuild allows the user to modify the voxel dimensions and rotate the volume by an arbitrary amount in any plane. This allows

Figure 6.9 Left image shows a patient scan with extremely bad noise, severely affecting the contours. After applying the smoothing filter twice, the contours in the right image more accurately reflect the desired structures

volume inspection and editing in any plane the user requires. The volume can also be resampled at a higher or lower resolution, to suite the needs of each individual case. This results in a finer or coarser approximation of the volume. By resampling at a lower resolution, processing time can be reduced for low-end systems. Resampling at a higher resolution can help improve the final biomodel quality of a low-quality scan.

6.2.6 Image processing

To aid general editing, BioBuild also supports linear Alters to smooth and sharpen images. Image smoothing can be extremely useful for the removal of noise from an image. Consider Figure 6.9. On the left is a patient scan with very bad scatter noise due to a hip replacement on the other non-visible side. On the right is the same scan, at the same threshold, after smoothing has been applied. Note that most of the scatter noise has been removed, leaving only the major structures.

This technique can be used to remove noise, as well as to remove small artefacts in an image series. Note that smoothing is applied across the entire data series, and not single images, to ensure consistency and correctness.

6.2.7 Build orientation optimization

The patented one-click build orientation optimization [27] is one of the strongest features of BioBuild, as the orientation of a biomodel has a large impact on both the time and cost of its production. This is because different orientations result in varying amounts of support materials and drastically altered build times. For example, on many RP systems, the time taken to build a biomodel can be significantly reduced if the model is built in an orientation that minimizes height. The one-click optimization allows the user to optimize the height of a build with a single click, reducing both the time to build and the support structures required.

Figure 6.10 BioBuild features a fully interactive 3D visualization window, which allows the user to rotate, pan and zoom the model

6.2.8 3D visualization

After all the necessary volume editing has been completed, the user should inspect the results in a 3D surface rendering. It is vitally important that the user inspects the rendered surface at the selected threshold to ensure the suitability of the threshold, and confirm the presence of all the structures required in the physical biomodel. This is achieved using the 3D visualization capabilities of BioBuild to create a 'virtual biomodel'. In the 3D view the user can fully interact in real time with the model by panning, zooming and rotating the model into any desired position. It is possible to zoom in for extreme close-ups of the finest features in the model, including internal structures, as well as view the model from any angle.

Figure 6.10 illustrates the 3D view. The image on the left is at the default zoom level, while the image on the right shows the same model, in the same orientation, at a much higher zoom factor. The model is easily manipulated using the standard three-button mouse interface of left button for rotation, middle button for pan and right button for zoom. There are also icons for immediately moving the model into predefined orientations, such as front, back, left and right. This viewing capability makes inspecting the 'virtual biomodel' quite simple.

When generating the surface model for the 3D view, the user has the ability to select a quality setting. This in no way affects the quality of the final biomodel, but it can reduce the complexity of the viewed surface model for systems without the necessary CPU or graphics performance.

6.2.9 RP file generation

To produce a physical biomodel it is necessary to generate a file specifying the biomodel surface in a format that an RP machine can interpret. BioBuild allows the user to generate an RP file at any time, either as an STL surface or as an SLC contour file.

STL file generation defaults to creating a surface mesh at a resolution defined by the coarsest voxel dimension. This is typically the Z plane or CT scan direction. Traditionally, spiral CT scans used are 1.0-1.5 mm apart. At this resolution, some interpolation is required to improve biomodel resolution and consequently surface finish. With submillimetre slice spacings becoming more prominent with the uptake of multislice CT systems, the user can now create RP build files with an STL facet size the same as the original scan spacing, requiring no image interpolation. Scan spacings of 0.3-0.5 mm over previously unachievable anatomical distances can produce excellent results in the resultant biomodels. Of course, attention then has to be paid to large STL file sizes and their effects on system performance in these instances. This is increasingly less important, however, as computing power continues to improve.

The surface model is generated from the current volume and threshold setting. Typically, the user will iterate between volume editing and 3D visualization several times before generating a STL surface file to ensure the best outcome.

A direct contour interface to RP via the SLC format still remains more optimal in terms of file size and processing requirements, but it lacks the portability of STL and requires separate build support structure software. A user creates an SLC file from the current volume with its current threshold by simply clicking the 'build contour' button and selecting the desired layer thickness. The layer thickness selected must then correspond directly to the build layer thickness of the target RP or 3D Printing hardware. The image volume will then be interpolated to a new volume at this selected thickness, and then contours will be extracted from each interpolated slice as per the set intensity threshold. The extracted contours will then reflect the anatomical structures of interest in each layer, as defined by the threshold.

It is also possible to select multiple volumes at the same time when generating an STL or SLC file, which allows for selective colouration (as available in SLA), or batch processing. This is useful when creating RP files with different thresholds set for each volume. Labels for biomodels are constructed in this fashion.

6.3 Future Enhancements

BioBuild is continually being updated to meet the demands of clients and to make the user interface more intuitive and even simpler. Part of this redesign includes functionality enhancements such as multiple simultaneous thresholds, each with its own colour code, advanced surface generation techniques [32], direct volume rendering (DVR) techniques [33, 34] for rapid biomodel preview without the need for surface generation and new powerful 3D editing tools. Further, with the increasing size of patient scans it is imperative that all operations remain interactive and that any dataset, regardless of size, be editable in an interactive manner. To this end, the memory model of BioBuild has been optimized for arbitrarily large volumes, and the software is incorporating the latest in DVR techniques, as well as making use of the recent improvements in commodity graphics cards [33].

6.3.1 Direct volume rendering (DVR)

The goal of DVR is to render a volume in real-time without the extraction of a surface. That is, the volume data are rendered directly. This is achieved by an optical data model that maps scalar values to optical properties like colour and opacity [34], commonly termed a transfer function. Typical approaches to DVR are based on multitexture rendering, in which the volume is represented by a series of images, blended together to give the appearance of a 3D object.

Until recently, real-time volume rendering was only achievable on high-end, expensive workstations. With recent major advances in commodity graphics cards, however, it is now possible to render very large datasets on inexpensive, consumer-grade hardware. This has opened the way for DVR on the desktop. This is important for biomodelling, as the trend towards larger and larger datasets means radiographers/technicians increasingly rely on volumetric reconstructions. Thus, it is imperative that BioBuild support DVR for volumetric preview and editing.

6.4 Conclusion

BioBuild has been found to be user-friendly, stable, efficient and accurate for biomodelling applications. It supports a wide variety of processing tools, including image filters, volume morphology and algebra, patented one-click orientation optimization and export to STL and SLC formats. The interface is simple and intuitive: once accustomed to the BioBuild paradigm, most operations become automatic. All this, combined with a design centred on supporting as wide a variety of scanners and data formats as possible, makes BioBuild the perfect bridge between patient scans and physical biomodels in almost any environment.

References

Eisenberg, R. (1992) Radiology: An Illustrated History, Mosby Year Book, Missouri.

Hemmy, D. C., David, D.J. and Herman, G.T. (1983) Three-dimensional reconstruction of craniofacial deformity using computed tomography. Neurosurgery, 13, 534-41. Marsh, J. L. and Vannier, M. W. (1983) Surface imaging from computerized tomographic scans. Surgery, August, 94(2), 159-65.

Vannier, M. W., Marsh, J. L., Gado, M. H., Totty, W. G., Gilula, L. A. and Evens, R. G. (1983) Clinical applications of three-dimensional surface reconstruction from CT scans: experience with 250 patient studies. Electromedica, 51, 122-31. Alberti, C. (1980) Three-dimensional CT and structure models. Br. J. Radiol. March, 53, 261-2. Blake, G. B., MacFarlane, M. R. and Hinton, J. W. (1990) Titanium in reconstructive surgery of the skull and face. Br. J. Plast. Surg., 43, 528-35. White, D. N. (1982) Method of forming implantable prostheses for reconstructive surgery. US Patent 4436683, June 3.

Zonneveld, F. W. and Noorman van der Dussen, M. F. (1992) Three-dimensional imaging and model fabrication in oral and maxillofacial surgery. Oral Maxillofac. Surg. Clin. N. Am., 4 (1), 19-33. Kalender, W. A., Seissler, W, Klotz, E. and Vock, P. (1990) Spiral volumetric CT with single breath hold technique, continuous transport, and continuous scanner rotation. Radiology, 176, 181-3. SME (1997) Rapid Prototyping Technology: a Unique Approach to the Diagnosis and Planning of

Medical Procedures, Society of Manufacturing Engineers, Dearborn, Michigan. Aoki, S., Sasaki, Y., Machida, T., Ohkubo, T., Minami, M. and Sasaki, Y. (1992) Cerebral aneurysms:

detection and delineation using 3-D-CT angiography. Am. J. Neuroradiol., 13, 1115-20. Goldwasser, S. M., Reynolds, R. A., Talton, D. A. and Walsh, E. S. (1988) Techniques for the rapid display and manipulation of 3-D biomedical data. Comp. Med. Imag. and Graphics, 12 (1), 1-24. Robb, R. A., and Hanson, D. (1991) A software system for interactive and quantitative visualization of multidimensional biomedical images. Aust. Phys. Eng. Sci. Med., 14 (1), 9-30. Drebin, R. A. (1988) Volume rendering. Computer Graphics, 22, 65-74.

Lorensen, W. E. and Cline, H. E. (1987) Marching cubes: a high resolution 3D surface construction algorithm. Computer Graphics, 21, 163-9. Hull, C. W. (1986) Apparatus for production of three-dimensional objects by stereolithography. US Patent 4575330, March 11.

Mankovich, N. J., Cheeseman, A. M. and Stoker, N. G. (1990) The display of three-dimensional anatomy with stereolithographic models. J. Digit. Imaging, 3, 200-03.

Klein, H. M., Scheider, W., Alzen, G., Voy, E. D. and Gunther, R. W. (1992) Pediatric craniofacial surgery: comparison of milling and stereolithography for 3D model manufacturing. Pediatr. Radiol., 22, 458-60.

Barker, T. M., Earwaker W. J. S., Frost, N. and Wakeley, G. (1993) Integration of 3-D medical imaging and rapid prototyping to create stereolithographic models. Aust. Phys. Eng. Sci. Med., 16, 79-85.

Robb, R. A. (1995) Three-Dimensional Biomedical Imaging - Principles and Practice, VCH Publishers, New York.

Barker, T. M., Earwaker, W. J. S. and Lisle, D. A. (1994) Accuracy of stereolithographic models of human anatomy. Australas. Radiol., 38, 106-11.

D'Urso, P. S. (1997) Stereolithographic biomodelling in Surgery. Unpublished PhD thesis, University of Queensland.

D'Urso, P. S., Barker, T. M., Earwaker, W. J. S., Arvier, J., Lanigan, M., Atkinson, R. L., Tomlinson, F. H., Loose, S., Donerly, W., Wilson, A., Weidmann, M. J., Redmond, M., Hall, B. I., Yau, Y. Y., Askin, G., Mason, S., Merry, G., Kelly, G., Frost, N., Wakeley, G. D., Riek, A. T., M'Kirdy, B. H., Stewart, J. and Effeney, D. J. (1994) The Australian SLA modelling project. Proceedings of the 2nd International Workshop on Rapid Prototyping in Medicine, April 7-8,1994, Zurich, Switzerland, p. 8.

D'Urso, P. S., Barker, T. M., Effeney, D. J. etal. (1993) Stereolithographic (SLA) biomodelling in surgery, preliminary results at 18 months. Proceedings of the Surgical Research Society of Australasia, Annual Scientific Meeting, July 21-23, 1993, Brisbane, Australia, p. 24.

Arvier, J. F., Barker, T. M., Yau, Y. Y., D'Urso, P. S., Atkinson, R. L. and McDermant, G. R. (1994) Maxillofacial biomodelling. Br. J. Oral Maxillofac. Surg., October, 32 (5), 276-83.

Yau, Y. Y., Arvier, J. F. and Barker, T. M. (1995) Technical note: maxillofacial biomodelling -preliminary result. Br. J. Radiol., May, 68 (809), 519-23.

D'Urso, P. S. (1993) Stereolithographic modelling process. US Patent 5 741 215, September.

D'Urso, P. S., Atkinson, R. L., Lanigan, M. W., Earwaker, W. J. S., Bruce, I. J., Holmes, A., Barker, T. M., Effeney, D. J. and Thompson, R. G. (1998) Stereolithographic (SL) biomodelling in craniofacial surgery. Br. J. Plast. Surg., 51 (7), 522-30.

D'Urso, P. S., Askin, G., Earwaker, W. J. S., Merry, G., Thompson, R. G., Barker, T. M. and Effeney, D. J. (1999) Spinal biomodelling. Spine, 24 (12), 1247-51.

D'Urso, P. S., Thompson, R. G., Atkinson, R. L., Weidmann, M. J., Redmond, M. J., Hall, B. I., Jeavons, S. J., Benson, M. D. and Earwaker, W. J. S. (1999) Cerebrovascular biomodelling. Surg. Neurology, 52 (5), 490-500.

D'Urso, P. S. and Thompson, R. G. (1998) Fetal biomodelling. Aust. and New Zealand J. Obstetrics and Gynaecol., 38, 205-7.

Klein, R. and Schilling, A. (1999) Fast distance field interpolation for reconstruction of surfaces from contours. In Eurographics '99, Short Papers and Demos Proceedings, Milan, Italy.

Rezk-Salama, C., Engel, K., Bauer, M., Greiner, G. and Ertl, T. (2000) Interactive volume rendering on standard PC graphics hardware using multi-textures and multi-stage rasterization. Proceedings of the ACM SIGGRAPH/EUROGRAPHICS Workshop on Graphics Hardware, 2000, Interlaken, Switzerland, pp. 109-18.

Engel, K. and Ertl, T. (2002) Interactive high quality volume rendering with flexible consumer graphics hardware. In Eurographics '02, State of the Art Report.

Was this article helpful?

0 0
Arthritis Relief and Prevention

Arthritis Relief and Prevention

This report may be oh so welcome especially if theres no doctor in the house Take Charge of Your Arthritis Now in less than 5-Minutes the time it takes to make an appointment with your healthcare provider Could you use some help understanding arthritis Maybe a little gentle, bedside manner in your battle for joint pain relief would be great Well, even if you are not sure if arthritis is the issue with you or your friend or loved one.

Get My Free Ebook


Post a comment