The Basics of Contrast-Enhanced CT
Intravenous contrast enhancement is fundamental to CT imaging. Here’s how it might be affecting your CT image datasets
The use of contrast is ubiquitous in radiologic imaging and yet, there’s surprisingly little helpful information on the topic online. Here, I’ll describe the basics of IV contrast, how it appears in CT images, and how it might affect your imaging dataset.
As a data scientist working on medical imaging studies, it is important to understand how contrast is being used in your datasets because contrast strongly affects the appearance of the CT exam, both intensity and texture. Depending on the application, contrast may be the main source of your imaging features, or it may be an unwanted source of variation. You will also find that the use of contrast can be subtle, and sometimes poorly labeled, so having a good working knowledge of this topic will help you organize your dataset and plan your study much more effectively.
Basics of CT attenuation
To understand why contrast agent is used in CT exams, you first must understand a bit about X-Ray physics. A CT scanner is a 3D imaging system that measures the X-Ray energy which is absorbed by organs and tissues in the body. Inside the scanner, a rotating X-Ray tube emits high-energy photons through the body and into a detector array, which measures the decay rate, or attenuation, of the photon energy.
Within the body, dense structures like cortical bone will block a greater proportion of photons, increasing the rate of attenuation. These high-attenuating regions will appear very bright in the resulting scan.
Less-dense structures like the air-filled lungs will pass photons and appear dark. Attenuation is reported in Hounsfield Units (HU) (named after the inventor of CT, Godfrey Hounsfield) and are normalized such that air has an attenuation of -1000HU and water has an attenuation of 0HU. Fat tissue has a typical attenuation of -100 to -50HU, blood has an attenuation of +30HU, and soft tissue has an attenuation of +10 to +50HU.
Contrast material
The purpose of contrast is to highlight tissues that otherwise would be difficult to distinguish from their surroundings. In the absence of contrast, most of the soft tissue in the body has a very similar X-Ray absorption, and will appear a uniform gray.
Contrast agent is typically either intravenous (injected into a vein), or ingested (swallowed), although contrast has been adapted for every conceivable method of administration. Ingested contrast is used to highlight the stomach, small, and large bowel in gastrointestinal imaging. Intravenous (IV) contrast, which is much more common, is used for a variety of indications (it can also be coupled with ingested contrast).
In both cases, the contrast agent consists of an inert heavy metal such as iodine or barium which is suspended in a liquid solution. The heavy metal has a very high attenuation which preferentially blocks X-Ray energy and appears bright in the resulting CT image.
The appearance of IV contrast in CT
Intravenous contrast captures movement of blood through the body. At the simplest level, tissues that have more blood will appear brighter, or more enhanced. In the abdomen CT exam below, the left CT was acquired before contrast was injected, and the right CT was acquired after. Adipose (fat) tissue, which is poorly perfused, has a similar appearance in both images, however the liver and spleen, which both receive a high perfusion, appear brighter (more enhanced) in the right-most image. Notice the biggest difference is in the aorta, and the vessels of the liver and spleen.
Another example is shown below, a postcontrast head CT from an individual who has suffered a hemorrhagic stroke. The bright pool of contrast in the scan indicates the location and severity of the bleed.
Contrast phase captures perfusion change over time
Circulation is a dynamic process. Consider a small injection of contrast agent into a vein in the arm (to ensure the timing of the bolus remains consistent, the contrast is pushed into the with a pump, sometimes at up to 300psi). That bolus of contrast will take a few seconds to travel up to the heart and lungs. From the heart, it will pass through the aorta, branching into the major arteries, then into the tissues of the body through the capillary system, finally returning back through venous circulation back to the heart. Over several cardiac cycles, the kidneys will start to filter contrast agent from the bloodstream, excreting it into the bladder.
CT contrast imaging leverages this dynamic aspect of circulation by capturing images at one or more specific points in time, or phases, after injection.
Hundreds of CT imaging protocols capture various combinations of these contrast phases. These phase-contrast CTs will be precisely timed to the moment of contrast injection, and the timing will vary based on the purpose of the scan and the preference of the radiologist ordering the test.
For example, contrast enhancement is particularly useful to image the kidneys, which have a characteristic pattern of perfusion. To capture this, imaging protocols are timed to capture the precise moments that contrast agent is passing through the kidneys. One publication I helped with as a graduate student was a validation of explainable AI techniques from a classifier trained on renal contrast CT.
As a general rule, across indications, CT contrast phase is simplified into 4 categories:
- Pre-contrast or Noncontrast, no IV contrast is present in the body (the exam may be acquired before contrast is injected, or contrast may not have been used at all).
- Arterial phase, typically 15–30s postinjection, contrast is perfusing through the heart, aorta, and arteries.
- Venous phase, typically ~60s postinjection, contrast is perfusing through the tissues and returning to circulation.
- Delayed phase or Nephrographic phase, typically 120–180s postinjection, contrast is in the process of being filtered by the kidneys and removed from circulation.
A CT study may have just one contrast phase, it may have all four phases, or any combination.
What you need to know for your data science project
First, be aware that the use of contrast will affect the appearance of nearly every tissue in the body besides fat tissue and free fluid; even tissues you may not expect, like bone, can vary in intensity by 20–30HU based on the use of contrast.
Combinations of contrast series will vary depending on the indication, and the preference of the radiologist who ordered the exam. Protocols are also periodically updated. This means that there is a high likelihood that if you are combining exams from different patient cohorts, different hospitals, or different scanners, you will have inconsistent combinations of contrast exams in your dataset.
There are two ways to handle this. If you are using the CT for its intended application (e.g., you are training a model on glioblastoma cases from head CTs), decide exactly which CT contrast phases you will include and enforce it as a selection criteria. For example, you may require that all glioblastoma cases must have BOTH precontrast and arterial phase contrast scan; if an exam doesn’t have that exact combination, it is excluded. Additional series are also excluded. This is typically a conversation I have with the radiologist when discussing study design.
If you are using the CT for a incidental application, you can usually be more lenient. For example, our bone-density model, which segments the spine, was trained on all four contrast phase images. If the patient has multiple contrast phase series, we simply average the bone measurements from all phases. We have done extensive validation demonstrating that IV contrast has no effect on model accuracy, and only a slight effect on subsequent density measurements (no more than 10% error) which we acknowledge as a limitation of our current approach.
Finally, and this is probably the most unfortunate fact, contrast phase is not generally well labeled. It is usually indicated in the series description in abbreviated form but these can be hard to interpret. There is a set of Tags in the DICOM header, starting with (0018, 0010) Contrast/Bolus Agent but I have found these to be inconsistently reported. The only reliable way I have found to accurately label contrast phase with 100% confidence is to read the images themselves.
CT contrast phase is one field that stands to benefit quite a bit from recent breakthroughs in deep learning. My colleague and good friend Gian Marco Conte has a publication in the journal Radiology where he trained a GAN to synthesize missing MRI sequences for glioblastoma, a similar approach could be used to synthesize CT contrast phase.
As a patient…
Although I’ve never had a CT, I’ve personally had several IV contrast MRIs done, and can reassure that the use of contrast is painless. An IV line is placed in your arm prior to the procedure, during the exam itself when the contrast is pushed, it feels like a cold rush to your arm. Some people claim that they get a metallic taste in their mouth, but I’ve never experienced this.
