The interpretation of CT colonography requires focused interactive training using specific computer software that allows dynamic viewing of two dimensional axial images, multi-planar reformats (MPR) and three-dimensional renderings. Readers may use either 2D axial images or 3D renderings for primary interpretation of CTC, with the alternate method reserved for problem solving specific questions related to a potential lesion. Current emphasis is for readers to develop skills for CTC interpretation using both two and three dimensional images in order to optimize lesion detection and to decrease interpretation times. Proficient utilization of these techniques, gained by appropriate training, has been shown to correlate with polyp detection sensitivity. The sensitivity for detection of polyps 6 mm or larger during the training period was found to be the sole independent predictor for subsequent sensitivity for similar lesion detection37
Primary 2D interpretation is performed by scrolling through magnified axial images of the colon obtained in supine and prone positions. CT window settings should be set to maximize detection of intraluminal lesions; high contrast colon window settings should approximate a window width of 1,500 and level of -300. Sessile polyps have round or ovoid morphology and are of soft tissue density. They remain fixed in location on the colon wall in both the supine and prone images. [FIGURE ] Stool can be differentiated from polyps since it is typically of mixed density and shifts location when the patient changes position. Pedunculated polyps can shift in location when the patient moves from supine to prone positions, but the stalk is typically easily identified on 2D and 3D images. [FIGURE ] Multiplanar reformats and 3D images are useful for evaluating lesion morphology and for confirming polyps. Primary 2D interpretation allows for rapid assessment of lesion density and homogeneity and has been shown to have overall relatively shorter interpretation times compared to primary 3D interpretation6
Sessile Polyp. Axial view (A) shows a sessile polyp in the ascending colon. The 3D endoluminal view (B) reveals a typical spherical appearance of a sessile polyp.
Pedunculated Polyp - Axial view (A) shows a large pedunculated polyp with a short stalk in the sigmoid colon. The 3D endoluminal view (B) shows the pedunculated polyp arising from a haustral fold and projecting into the lumen.
Primary 3D interpretation is performed by endoluminal navigation through the entire colon lumen along a computer-generated center line. Anterograde and retrograde navigations in both supine and prone positions are required in order to assure complete assessment of the colon wall. Evaluation during both retrograde and anterograde navigations allows for visualization of both sides of colonic folds. Similarly, layering fluid or residual stool could obscure polyps when viewing 3D navigations in only one position. When a potential lesion is identified on the 3D view, the reader places a bookmark near the lesion and then evaluates the lesion density using the axial or multiplanar views. Lesion mobility should then be assessed by comparing the lesion location and orientation from supine to prone positions. Additional CTC software algorithms are available to improve discrimination of true polyps from pitfalls. A color density map placed on the 3D endoluminal view aids in differentiating densely tagged stool from the soft tissue density of polyps without having to refer back to the 2D display. Another algorithm paints the viewed surfaces of the colon on the 3D endoluminal view a different color and then presents a list of patches of wall that have not been previously viewed so that the reader can rapidly perform a complete review of all surfaces.
The primary 3D interpretation technique may be easier for some inexperienced readers to learn. Additionally polyps on or near colon folds are often seen better on 3D views. Interpretation times for primary 3D reads tend to be longer than for primary 2D reads. Comparison of primary 2D versus primary 3D interpretation methods was performed as a part of the ACRIN CTC trial38
. The sensitivity and specificity for primary 2D readers was 0.84 and 0.86 respectively for polyps 10mm or larger. Readers who employed both 2D and 3D techniques demonstrated a sensitivity and specificity of 0.84 and 0.83 respectively. The sensitivity and specificity of primary 3D readers was 0.76 and 0.82 respectively. There was no statistically significant difference in performance between any of these groups.
Various CT colonography techniques are being evaluated in an attempt to improve accuracy and decrease reading times. The virtual dissection view is a novel display in which the tubular shaped colon is bisected through its long axis and then opened and flattened for an internal display. The dissection view allows visualization of larger portions of the colon wall simultaneously and may help to shorten interpretation times. However, there may be distortion of normal folds and of lesions with the flattening technique so readers must be specifically trained to interpret these views.
Electronic cleansing is a post-processing technique involving application of computer algorithms to subtract intra-colonic high-density tagged material. Residual material in the colon is tagged using ingested barium and/or iodinated contrast administered at specified time periods on the day prior to the CTC. Following electronic cleansing and removal of the tagged stool and fluid, soft tissue polyps and cancers are better visualized. The need for laxative cleansing of the colon may potentially be reduced or eliminated with optimum tagging of residual material combined with electronic cleansing. The effect of electronic subtraction is limited in the presence of partial volume artifacts and heterogeneous tagging of stool. Another potential pitfall of electronic cleansing is “over-subtraction,” where parts of soft tissue density polyps are removed along with tagged material causing incorrect sizing or false negatives. A study of 114 patients undergoing tagged, non-cathartic CT colonography showed that stool subtraction increased sensitivity for the detection of polyps larger than 1 cm and polyps between 6 to 9 mm, compared to non-subtracted images39
. A prospective evaluation of 605 asymptomatic patients evaluating the diagnostic accuracy of laxative-free CTC found that the per-patient sensitivity for the detection of adenomas 10 mm or larger, 8 mm or larger and those 6 mm or larger were 91%, 70%, and 59% respectively40
. A low osmolar, nonionic iodinated contrast agent was used for tagging.
Computer aided detection (CAD) for CT colonography uses computer algorithms based on polyp features such as shape to identify potential lesions. These computerized methods may be used to improve sensitivity for polyp detection particularly in novice readers and to reduce inter-reader variability. CAD may employed as a primary reader to identify possible polyps and cancers prior to the radiologist's interpretation or as a secondary check after the initial unassisted interpretation by the radiologist in order to confirm findings and identify any additional missed lesions. CAD may also be used as a concurrent reader, in which a radiologist reads the CTC images while assisted by CAD with potential lesions identified as targets by the software. CAD is most commonly used as a secondary reader. A study comparing CTC interpretation using CAD as a secondary versus concurrent reader found that CAD functioning as a secondary reader significantly improved reader sensitivity but use of CAD concurrently had no effect41
. Neither concurrent nor secondary CAD improved the specificity of CTC. Increased sensitivity for polyp detection using CAD must be offset by adequate specificity so that false positives are minimized and interpretation times are acceptable. A study comparing CAD applied to both tagged and untagged CTC examinations found that there were no significant performance differences with similar sensitivity of about 90% for polyps 6 mm or larger42
. Detection rates were demonstrated to be comparable to human readers at a relatively low false-positive rate of about 5 per patient dataset.
The CT colonography Reporting and Data System (C-RADS) was developed to promote more consistent and clear communication of CT colonography results43
. Lesion descriptors including size, morphology, segmental location and density are recommended to assist in standardizing CTC reports. The classification system consists of categories ranging from inadequate study (C0) to likely malignant colonic mass (C4).