It is important to understand that sonologists around the world have always conjured three-dimensional images of anatomy or pathology in their minds while doing their 2D scans. However, until recently it was not possible to do this type of reconstruction on patient data acquired using ultrasound. The advent of 3D and 4D ultrasound for the first time allowed us a peek into the mechanics of thinking of a sonologist by acquiring the volume data and allowing us to reconstruct the images in different planes on the ultrasound machine or a workstation. The problem is that we are not used to manipulating this data in such an “external” visualization format. It takes some understanding of the basics of volume acquisition and manipulation to understand the dynamics of reconstruction. It is perhaps easier for radiologists who have dealt with axial imaging in CT and multiplanar imaging in MRI since they are used to sectional imaging. Additionally the ability to render the volume using different algorithms makes it seem more complicated. However, the basic steps are simple, and a logical approach to 3D and 4D scanning can help in its complete comprehension, and even makes it easier to scan in many instances. Remember, multiplanar imaging is the greatest benefit of Volume scans.

3D ultrasound is a data set that contains a large number of 2D planes (B-mode images). This is important to understand. If the page of a book is one 2D plane, then the book itself is the entire data set. Once the Volume is acquired using a dedicated 3D probe you can “Walk” through the volume in a manner similar to leafing through the pages of a book, meaning you can walk through the various 2D planes that make up the entire volume. This is also known as translation. If conceptualizing a book is difficult you can even think of a stack of books in a bookrack – each book represents a 2D plane and each shelf on the bookrack is a volume. Once you have the data you can also flip it to see it in different planes by rotation. You have to remember that the probes usually acquire the data by moving a B mode transducer within a housing like a hand held Japanese fan (Fig 1).

Figure 1 - B Mode transducer in a housing that swivels on a motor in fan like motion.

In such a case the acquired volume unlike the defined rectangle shape of a book looks like a pyramid or triangle of volume information with a broad base (Fig 2)

This Volume shape can then be dissected in any plane, to get what we have been talking of as multiplanar imaging. It is like taking a block of un-sliced cheese and cutting it any plane you want.

4D ultrasound is also known as "Real-time 3D Ultrasound" – The basic concept being that the processing powers of the computers has increased so much that today we can get our ultrasound equipment to acquire and display the 3D datasets with their multiplanar reformations and renderings in real time, as we scan the patient. This has many benefits in obstetrical scanning where it is critical to analyze anatomy as it continues its physiological movements.

We have to decide on few things before we start a Volume acquisition.

1. Getting a good B mode image
2. The region of interest
3. The angle of the acquired volume
4. The quality of acquisition (this dictates time of acquisition)
5. The primary plane of acquisition – longitudinal or transverse
6. 3D or 4D

The region of interest should be as small as possible unless you want to look at a very large anatomy correlation. The angle of acquisition defines the swing/sweep of the probe and a wide angle will take longer acquisition time usually. The quality of the acquisition can be increased in a stepladder fashion up to “Maximum” – the better the quality selected the longer is the acquisition time. The reason is simple – a better quality usually needs more image lines in the acquisition plane and more slices in the volume. So if you are looking at scanning a crying baby you will have to scan at lower quality to complete your acquisition faster, compromising somewhat, on the resolution of your data. The multiplanar view that you wish to have in the best resolution should be tried to be the primary scan plane of acquisition – although this may not be always possible, it is important to be kept in mind (Fig 3).

Lastly, you need to decide before you start the acquisition whether you wish to have a 3D scan and work with the volume later on or whether you wish to have a live real time display of all planes and rendering by opting for 4D scan. A very thin volume 4D scan with some minor differences in mechanics of acquisition is alternatively called as VCI (volume contrast imaging) and can also be pre-selected to get the desired plane in real time. VCI increases the inherent contrast in the image and delineates borders in a better manner.

Figure 3 A: Defining the region of interest (as small as needed). B: knowing the Volume angle (on your monitor screen) C: Controls on the Panel telling Quality and Volume Angle

Once the automatic volume scan is initiated the hand must remain very still and the patient can be asked to hold breath briefly. This helps in reducing the breathing artifacts and gives better results. The acquisition is in a "fan" like manner (Fig 4)

A basic 3D image after acquisition looks like this green "volume block" illustration and it is possible to "translate" or "rotate" through this information (Fig 5 a,b).

Figure 5a - Translation of one of the planes from one side of the volume block to another.

Figure 5b - Rotation of a plane into any of the axis in the volume block

Once we have the volume, we must make best use of resources available to us to get the maximal information from our data. The key lies in use of various reconstruction and rendering algorithms. Additionally we have a number of gray maps and contrast adjustments with color available to apply and see if they can increase the yield of information. The system always allows two rendering modes – the types and the percentages used in the mixture of the modes depend on the user.

The basic rendering algorithms are

1. Surface Smooth
2. Surface Texture
3. Transparency maximum
4. X ray Mode
5. Gradient light/Light Mode
6. Transparency minimum

We can mix and match any of the two algorithms, and then decide the percentage of mixes. Its like adding two colors e.g. Red and white, and then altering the percentage of red or white, to get a darker or lighter pink shade. Similarly, a surface texture and a surface smooth can be mixed to get an image that is having surface texture but the edges are smoothened. A mixture of surface texture with gradient light or transparency minimum can also be done. These various permutations and combinations can initially be experimented with and then you can find your comfort levels with regards to the rendering mixes you wish to use. Some general rules are present that can be used for as initial guidelines – minimum transparency for cystic structures, maximum for bony renderings, surface texturing for soft tissues etc. We digress here a bit to look at what is volume rendering.

What is volume rendering?
Volume is a three-dimensional array of voxels. Just the same way an image is a 2D array of pixels. Voxel is the basic element of the volume. Typical volume size may be 1283 voxels, but any other size is acceptable. Volume Rendering means rendering the voxel-based data into viewable 2D image. There are two categories of volume rendering algorithms:
Ray casting algorithms (Object Order)

• Basic ray-casting
• Using octrees

• Basic slicing
• Shear-Warp factorization
• Transparent textures

Ray casting (projection ways) is a method in which for every pixel in the image, a ray is cast through the volume (Fig 6). The ray intersects a line of voxels. For every pixel in the output image, a ray is shot into the data volume. At a predetermined number of evenly spaced locations along the ray the color and opacity values are obtained by interpolation. The interpolated colors and opacities are then merged with each other and with the background by compositing in either front-to-back or back-to-front order to yield the color of the pixel (The order of compositing makes no difference in the final output image). The render algorithm finally determines how the structures are visualized.

Apart from the basic translation and rotation through the acquired volume, it is important to understand that correlation methodology can be used to zero in on a particular spot in any of the planes in an acquired volume. For this usually it is good to begin at the level of sectional planes. There is a small yellow dot that is seen in all the planes. This dot correlates a point of focus in all the planes. It is the same identical voxel seen from different directions. If you move this point and place it on a structure or a lesion in plane A then the point automatically moves and finds the corresponding structure or lesion location in the planes B and C. This is very important for localization (Fig 9).

In the rendered 3D image too the scan plane lines can be activated by accepting the ROI and the intersection of two scan plane lines usually is the yellow point of correlation. This is also known as Localization dot. (Fig 10)

Figure 10 - Correlation of a point on the rendered image using lines.

However in the rendered image you have to remember that you are looking at the volume and the point actually may not be on the surface although the lines will be seen intersecting on the surface. Also, in the rendered image you can get the representative lines from two planes – the third plane being end on will not be localized by the line.

This is the summation of all the information we have acquired into a proper and coherent understanding. It is also the most important component of the entire exercise. Finally it also helps in choosing the right image or reconstruction to display or store or record for the patient and the referring physician.

A recollection of all the tools available to you for interpretation of the data is warranted at this point. All these can help in understanding and analysis.

1. Translation – leafing through the 2D planes in a volume
2. Rotation – turning the planes to improve orientation up down, right left
3. Point Correlation - using the yellow dot to spot a lesion or pathology in all the corresponding planes.
4. Rendering direction – using one of the six manners available (top-bottom, front-back, etc) to look at the rendered volume from one of the sides
5. Rendering algorithms – surface texture, maximum, minimum modes etc. It is important to mix the two modes chosen in a acceptable manner.
Additionally you have three more tools to help in fine-tuning of your volume data and extracting more information -
6. Threshold
7. Magic Cut
8. Vocal – Volume Calculation

Threshold parameter can contribute significantly to quality of the rendered images. A threshold parameter once set can subdue the voxels with gray values below the threshold value (Fig 11). This can help in clearing up noise sometimes and giving further clarity to the image. Very high thresholds can cut-off important information so it must be used judiciously.

Figure 11 - Different threshold settings for a surface rendered image showing the lumen of the gall bladder. C and D are too high and too low, while A and B are just about adequate with A showing the details a bit better.

Magic cut can also be called as an electronic scalpel. It is used for “sculpting” the image. In fact working with the magic cut can many times increase your own understanding of how the 3D volume data set works. It helps in slicing away information that you do not want so that the information inside the volume or behind an obstruction can be visualized (fig 12). The depth at which magic cut should cut can be defined as full (the cut is through the entire depth of the volume) or defined (where you can manually select a partial depth of variable thickness). Sometimes selecting defined magic cut is better since we do not cut away all the information.

Figure 12 - A shows a 2D image of an ovarian cyst that was rendered by 3D using surface texture mode in B and C. The rendered cyst wall precluded a good view of the mural nodule seen on the 2D. Magic cut was used to slice away the cyst wall to show a clear view of the mural volume in the rendered image in D.

VOCAL™
This is a software program that allows us to draw contour of a region of interest, rotate that area over 180 degrees by predefined steps – and then calculate the volume of that selected region/lesion(Fig13). VOCAL™ measurements are performed by integration of polygon areas generated by rotation via a fixed axis (rotation axis of contours). The method used for the integration of the polygon areas is given by a complicated mathematical formula beyond the scope of this tutorial. This method of volume calculation is more sensitive than the conventional 2D method of calculating volumes using oblate spheroid formula: Volume = LxALxW/1.57.

The inter-operator variability is lesser that in the 2D formula based calculations.

Figure 13 - The contours of prostate are traced and using VOCAL™ volume is calculated – In this case the volume is 24.9cc.

Zoom:
A large number of users under-use the "zoom" facility. If the zoom is used while the rendering is active it can help in enlarging the region of interest. If the zoom is used after the rendering has been frozen it can still help in giving us an enlarged version of the pathology for good prints. Remember ultrasound imaging is difficult for the referring physicians to understand unless we are able to display the pathology in as lucid manner as possible. The enlarged image allows the physician to appreciate the pathology and focus on it rather than getting lost in the myriad of information.

3D Cine rotation.
The perception of depth is best seen while the 3D rendered image rotates on an axis. This is called the Cine mode. The rendering improves and the textures appear to be more realistic. In some views the noise actually decreases and the contours are better visualized. This can be even stored as an avi. loop on a CD or a DVD and given to the patient or sent to the referring physician as a record.

Finally it must be remembered that 3D or 4D cannot see what the B mode cannot see. This is critical to understanding that we are not talking in terms of 3D or 4D being better than B mode, or 3D/4D having the capability to ride over the B mode artifacts. The problems of B mode imaging are always carried into 3D/4D imaging. If shadowing from a calculus or air does not allow us a good acoustic window on 2D imaging then 3D cannot resolve that issue. What 3D/4D does, is that it allows us a better visualization of spatial relationships by multiplanar imaging, allows rendering abilities to convey information in a different manner, and for the first time actually allows storage of ultrasound data for review at a remote time or distance. The advancement in 3D and 4D technology has led to the development of 3D centric applications like STIC ( Spatial Temporal Image Correlation) to study the fetal heart, and advanced software applications like Inversion sequence etc. These will form the basis for an advanced tutorial on volume ultrasound.

1. Nelson T.R., Downey D.B., Pretorius D.H. et al. Three-dimensional ultrasound. Philadelphia: Lippincott Williams & Wilkins, 1999.
2. Rankin R.N., Fenster A., Downey D.B. et al. Three-dimensional sonographic reconstruction: techniques and diagnostic applications. // Amer. J. Roentgen., 1993. V.161. p. 695-702.
3. Rose SC. Related Articles, Links Paradigm shifts provided by a new means of seeing: the educational value of three-dimensional ultrasonography. J Ultrasound Med. 2001 Sep;20(9):937-40.
4. Yen JT, Smith SW. Real-time rectilinear volumetric imaging using a periodic array. Ultrasound Med Biol. 2002 Jul;28(7):923-31.
5. Khurana A, Dahiya N. "3D and 4D Ultrasound: A Text and Atlas." Anshan, U.K. 2004.
6. K. H. Höhne, H. Fuchs, S. M. Pizer"3D Imaging in Medicine Algorithms, Systems, Applications", Springer-Verlag
7. G. Sakas, P. Shirley, S. Müller "Photorealistic Rendering Techniques", Springer-Verlag
8. http://www.cs.wpi.edu/~matt/courses/cs563/talks/powwie/p1/ray-cast.htm

• Take Quiz