CT Colonography
 

1. Computerized Detection of Polyps   Colon cancer is the second leading cause of cancer deaths among men and women in the United States. Most colon cancers can be prevented if precursor colonic polyps are detected and removed. CT colonography, also known as virtual colonoscopy, has been proposed as a promising, non-invasive technique for colon cancer screening. However, in order for CT colonography to be deemed a practical screening tool, a large number of images must be interpreted in a prompt and accurate fashion. To this end, we have been developing a CAD scheme for the automated detection of polyps in CT colonography. This scheme provides radiologists with the location of suspected polyps, thereby allowing for quick and accurate detection of suspicious lesions. In this scheme, first, the colonic wall is extracted from an isotropic volume data set obtained by interpolation of CT colonography images along the axial direction. Polyp candidates are detected by computation of 3D geometric features that characterize polyps, followed by extraction of polyps with hysteresis thresholding and fuzzy merging by use of these geometric features. The regions of the polyp candidates are segmented by use of conditional morphological dilation. False-positive polyp candidates are reduced by a Bayesian neural network with shape and texture features. This scheme was applied to 121 CT colonography cases, 28 of which contained a total of 42 colonoscopy-confirmed polyps. In both by-patient and by-polyp analyses, the sensitivity of our scheme was 93%, with an average false-positive rate of 2.0 per patient, by use of a leave-one-out evaluation with by-patient elimination. The results indicate that the CAD scheme may be useful in improving the performance of computer-aided detection for colon cancer in a clinical screening setting. (550, 685, 686, 687, 772, 776, 788, 789, 824, 827, 837, 838, 842, 843, 854, 855, 856, 889, 957, 958, 959)	Volumetric shape index feature for differentiating polyps from normal structures
2. Computerized Detection of Masses in the Colon The automated detection of masses naturally complements the automated detection of polyps in CT colonography and would produce a more comprehensive computer aid to radiologists. We have been developing a scheme for the computerized detection of masses that can be integrated into the above CAD scheme for the detection of polyps. Two methods, fuzzy merging and wall-thickening analysis, were developed for the detection of masses. The fuzzy merging method detected masses with a significant intraluminal component by separating the initial CAD detections of locally cap-like shapes within the colonic wall into mass candidates and polyp candidates. The wall-thickening analysis detected non-intraluminal masses by searching the colonic wall for abnormal thickening. The final regions of the mass candidates were extracted by use of a level set method based on a fast marching algorithm. False-positive detections were reduced by a quadratic discriminant classifier. The performance of the scheme was evaluated by use of a leave-one-out method with by-patient elimination. A database of 82 CT colonography cases was used. 14 patients (17%) had a total of 14 masses of 30–50 mm, and sixteen patients (20%) had a total of 30 polyps 5–25 mm in diameter. The fuzzy merging method detected 11 of the masses, and the wall-thickening analysis detected 3 of the masses including all non-intraluminal masses. In combination, the two methods detected 13 of the 14 masses with 0.21 FPs per patient on average based on the leave-one-out evaluation. The results indicate that the scheme is potentially useful in providing a high-performance CAD scheme for the detection of colorectal neoplasms in CT colonography. (878, 936)
3.  Fully Automated Segmentation of the Colon The task of automated segmenting (or extracting) the colon is generally the important first step for automated detection of polyps.  We have developed several fully automated methods for segmentation of the colon in CT colonography.  One of the methods, called knowledge-guided segmentation (KGS), uses the anatomic knowledge of the abdomen for the extraction task. In this method, first, an isotropic 3D volume is generated from the transverse CT images by interpolation of the CT images along the transverse direction.  Then the method extracts the visible colonic wall by removing the normal structures that are not connected to the colon, based on thresholding of the 3D volume with the CT values characteristic of these structures.  The second step removes extra-colonic components that are connected to the outer surface of the colonic wall, such as the stomach and small bowel, by applying a self-adjusting volume-growing technique to the colonic lumen.  On average, the regions extracted by KGS can cover more than 98% of the visible surface region of the colonic wall. An advanced method, called centerline-based segmentation, has been developed for further improvement and speeding up of the colon segmentation process. The algorithm thresholds a set of unprocessed CT slices. Outer air is removed, after which a centerline is computed for all remaining regions in the thresholded volume, disregarding segments related to extra-colonic structures. Centerline segments are connected, after which anatomy-based removal of segments representing extra-colonic structures occurs. Segments related to the remaining centerline are locally region-grown, after which the colonic wall is found by dilation. Shape-based interpolation provides an isotropic mask. In an analysis of 38 CT colonographic cases, the extracted region covered approximately 96% of the colonic wall, with only 1% of the extracted region included extra-colonic structures.  (772, 825, 826, 937)	Automated segmentation of colon
4. Anatomic Correspondence between Supine and Prone Datasets   Radiologists often compare the supine and prone datasets of a patient to confirm potential polyp findings in CT colonography. We have developed a new automated region-based supine-prone correspondence method for the reduction of false-positive polyp candidates in CAD for CTC. Six anatomic landmarks are established by use of the extracted region of the colonic lumen. A region-growing scheme with distance calculations is used to divide the colonic lumen into overlapping segments that match in the supine and prone data sets. Polyp candidates detected by the CAD scheme are eliminated in colonic segments that have sufficient diagnostic quality and contain polyp candidates in only one of the data sets of the patient. The method was evaluated with 121 CTC cases, including 42 polyps in 28 patients. The cases were obtained by use of single- and multi-detector CT scanners with standard pre-colonoscopy cleansing. A complete or partial correspondence could be established in 71% of the cases. Based on a leave-one-patient-out evaluation, the application of the method reduced 19% of false positives reported by our CAD scheme at a 90.5% by-polyp detection sensitivity, without loss of any true positives. The resulting CAD scheme yielded 2.4 false positives per patient on average with the use of the correspondence method, whereas it yielded 3.0 false positives per patient without the use of the method. The results indicate that the correspondence method is potentially useful in improving the specificity of CAD in CTC.  (937, 938)	Automated supine-prone correspondence: numbers indicate CAD-identified landmarks and the black region exemplifies a matching region used for FP reduction
5. Virtual Endoscopic Visualization by Shape-Scale Signatures We have developed a new visualization method for virtual endoscopic examination of CT colonographic data by use of shape-scale analysis. The method provides each colonic structure of interest with a unique color, thereby facilitating rapid diagnosis of the colon. Two shape features, called the local shape index and curvedness, are used for defining the shape-scale spectrum. When we map the shape index and curvedness values within CT colonographic data to the shape-scale spectrum, specific types of colonic structures are represented by unique characteristic signatures in the spectrum. The characteristic signatures of specific types of lesions can be determined by use of computer-simulated lesions or by use of clinical data sets subjected to a computerized detection scheme. The signatures are used for defining a 2-D color map by assignment of a unique color to each signature region. The method was evaluated visually by use of computer-simulated lesions and clinical CT colonographic data sets, as well as by an evaluation of the human observer performance in the detection of polyps without and with the use of the color maps. The results indicate that the coloring of the colon yielded by the shape-scale color maps can be used for differentiating among the chosen colonic structures. Moreover, the results indicate that the use of the shape-scale color maps can improve the performance of radiologists in the detection of polyps in CT colonography. (939)	CAD workstation for detection of polyps in CTC
6. Fast and Robust Computation of Colon Centerlines We have developed a method for computation of a colon centerline that is fast and robust to the quality of the colonic lumen extracted from CT colonoscopic data. The proposed method first extracts local maxima in a distance map of a segmented colonic lumen. The maxima are considered to be nodes in a set of graphs, and are iteratively linked together, based on a set of connection criteria, giving a minimum distance spanning tree. The connection criteria are computed from the distance from object boundary, the Euclidean distance between nodes and the voxel values on the pathway between pairs of nodes. After the last iteration, redundant branches are removed and end segments are recovered for each remaining graph. A subset of the initial maxima is used for distinguishing between the colon and non-colonic centerline segments among the set of graphs, giving the final centerline representation. By use of 40 CT colonographic scans, the computer-generated centerlines, when compared with human-generated centerlines, had approximately the same displacement as when the human-generated centerlines were compared among each other (3.8 mm versus 4.0 mm). The coverage of the computer-generated centerlines was slightly less than that of the human-generated centerlines (92% versus 94%). The 40 centerlines were, on average, computed in 10.5 seconds, including computation time for the distance transform as compared with 12-17 seconds or more per centerline as reported in other studies. (864, 917)
7. Effect of CAD on Observers’ Performance in Detection of Polyps We conducted an observer study to evaluate the effect of CAD on human readers’ performance in the detection of polyps in CT colonography. Twenty data sets including 11 polyps were retrospectively selected from our CTC database. A sequential test was used, in which four observers interpreted CTC examinations by use of our colon CAD workstation in two sequential sessions: one without CAD and the other with CAD. At each session, the observers rated the confidence level regarding whether the case was abnormal, i.e., at least one polyp > 5 mm was present in the colon. Receiver operating characteristic (ROC) analysis was performed based on the confidence levels for the abnormality of the CTC data sets, and the area under the ROC curve (Az) was calculated as a measure of the observers’ detection performance. Results showed that, for all of the observers, the detection performance increased by use of CAD. The average Az value without and with CAD were 0.70 and 0.85, respectively. A two-tailed t-test showed that the difference between these Az values was statistically significant (p = 0.025). The result indicates that CAD can be a useful tool for improving human readers’ performance in the detection of polyps in CTC. (942)