Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

DOI: 10.1148/radiol.2451061116
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Baker, M. E.
Right arrow Articles by Macari, M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Baker, M. E.
Right arrow Articles by Macari, M.
(Radiology 2007;245:140-149.)
© RSNA, 2007

Gastrointestinal Imaging

Computer-aided Detection of Colorectal Polyps: Can It Improve Sensitivity of Less-Experienced Readers? Preliminary Findings1

Mark E. Baker, MD, Luca Bogoni, PhD, Nancy A. Obuchowski, PhD, Chandra Dass, MD, Renee M. Kendzierski, DO, Erick M. Remer, MD, David M. Einstein, MD, Pascal Cathier, PhD, Anna Jerebko, PhD, Sarang Lakare, PhD, Andrew Blum, MD, Dina F. Caroline, MD, PhD, and Michael Macari, MD

1 From the Departments of Radiology and Quantitative Health Sciences, the Cleveland Clinic Foundation, 9500 Euclid Ave, Hb6, Cleveland, OH 44195 (M.E.B., N.A.O., E.M.R., D.M.E.); Department of Radiology, Temple University, Philadelphia, Pa (C.D., R.M.K., A.B., D.F.C.); Department of Radiology, New York University School of Medicine, New York, NY (M.M.); and Siemens Medical Systems, Malvern, Pa (L.B., P.C., A.J., S.L.). From the 2005 RSNA Annual Meeting. Received June 29, 2006; revision requested September 1; revision received October 25; accepted November 22; final version accepted April 2, 2007. Address correspondence to M.E.B. (e-mail: bakerm{at}ccf.org).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Purpose: To determine whether computer-aided detection (CAD) applied to computed tomographic (CT) colonography can help improve sensitivity of polyp detection by less-experienced radiologist readers, with colonoscopy or consensus used as the reference standard.

Materials and Methods: The release of the CT colonographic studies was approved by the individual institutional review boards of each institution. Institutions from the United States were HIPAA compliant. Written informed consent was waived at all institutions. The CT colonographic studies in 30 patients from six institutions were collected; 24 images depicted at least one confirmed polyp 6 mm or larger (39 total polyps) and six depicted no polyps. By using an investigational software package, seven less-experienced readers from two institutions evaluated the CT colonographic images and marked or scored polyps by using a five-point scale before and after CAD. The time needed to interpret the CT colonographic findings without CAD and then to re-evaluate them with CAD was recorded. For each reader, the McNemar test, adjusted for clustered data, was used to compare sensitivities for readers without and with CAD; a Wilcoxon signed-rank test was used to analyze the number of false-positive results per patient.

Results: The average sensitivity of the seven readers for polyp detection was significantly improved with CAD—from 0.810 to 0.908 (P = .0152). The number of false-positive results per patient without and with CAD increased from 0.70 to 0.96 (95% confidence interval for the increase: –0.39, 0.91). The mean total time for the readings was 17 minutes 54 seconds; for interpretation of CT colonographic findings alone, the mean time was 14 minutes 16 seconds; and for review of CAD findings, the mean time was 3 minutes 38 seconds.

Conclusion: Results of this feasibility study suggest that CAD for CT colonography significantly improves per-polyp detection for less-experienced readers.

© RSNA, 2007


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Computed tomographic (CT) colonography is an evolving method for detecting colon polyps (1–13); only a few centers receive reimbursement for screening studies (14). In one large study, Pickhardt et al (6) suggests that CT colonography is as sensitive as optical colonography in the detection of adenomas, whereas other reports show much less promising results (7,12). However, there are substantial differences between these two studies in terms of method used to interpret CT colonographic findings, number of institutions, number of readers, and other variables. Hence, it is hard to properly compare the reported performance. Largely as a result of the uncertain sensitivity of polyp detection, the nonradiologic medical community has not yet fully embraced CT colonography as a reasonable alternative to optical colonography (13,15,16).

The CT colonographic examination demands time and effort on the part of a radiologist. Furthermore, there is a steep learning curve for the reader to attain competence, especially with two-dimensional image interpretations (17). Training programs have been developed by specialty centers (17,18); however, even after training, without continual feedback from a research protocol or optical colonographic results, radiologists are unlikely to know their sensitivity or false-positive rate. Additionally, there are many causes for interpretative errors in CT colonography (19,20), often exacerbated by reader fatigue.

Computer-aided detection (CAD) has improved the identification of breast cancer at mammography (21–24) and has potential in aiding detection of pulmonary nodules at CT for lung cancer screening (25–28). Thus, the purpose of our study was to determine whether CAD applied to CT colonography can help improve sensitivity of polyp detection by less-experienced radiologist readers, with colonoscopy or reader consensus used as the reference standard.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Siemens Medical Solutions (Malvern, Pa) provided the equipment (hardware and software) and technical assistance during the study. The authors who were not employees of or consultants for, or had not received support from, Siemens Medical Solutions had control of inclusion of any data and information that might present a conflict of interest for those authors who are employees of or consultants for, or had received research support from, Siemens Medical Solutions.

CAD Algorithm
The CAD (ColonCAD-2005 prototype VA10A; Siemens Medical Solutions) algorithm (29) used in our study included the following phases: data preprocessing, candidate generation, feature extractions, and pruning or filtering. During the data preprocessing phase, the colon is segmented; in the candidate generation phase, loci of detection are identified. The purpose of the candidate generation phase is to achieve high sensitivity. Thus, a large number of candidates may be accrued. These candidates are sequentially processed during the feature extraction phase. The features are based on tissue intensity, volumetric and surface shape, and texture characteristics. Each candidate, uniquely identified, and the associated features are then fed to a classifier—the pruning or filtering phase (30). Candidates are then evaluated, and features are labeled as potential lesions; candidates are presented to the reader once the reader requests to review the CAD findings. The labeling (or classification) is performed by a classifier that has been trained off-line from a training data set and then frozen for use in the CAD system. The critical requirement of a CAD system, and thus of its classifier, is its ability to generalize well. Namely, it should correctly label new data sets. The CAD findings were displayed on a single monitor in the transverse, coronal, simulated, air-contrast barium enema, and endoluminal views (Fig 1). The CAD system used in this study was developed in early 2004, and its sensitivity and number of reported false-positive results have since improved, as demonstrated in subsequent reader studies (31,32).


Figure 1
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: Flow diagram shows total CT colonographic population used for the investigation and for partitioning the CT studies into groups for CAD training, physician demonstration and training, and testing of CAD. Target condition present = at least one polyp larger than 6 mm.

 
CT Colonographic Studies Used
A database of 260 cases of thin-section CT colonographic examinations was contributed by six institutions (Fig 1): two from the United States (New York University Medical Center, New York, NY, and Cleveland Clinic Foundation, Cleveland, Ohio) and four from Europe (Vienna Hospital, Vienna, Austria; Notre Dame Hospital, Tournai, Belgium; La Sapienza Hospital, Rome, Italy; and Muenster Hospital, Muenster, Germany). The release of these cases was approved by the individual institutional review boards of each site. The U.S. institutions complied with provisions of the Health Insurance Portability and Accountability Act. Written informed consent was waived at all institutions. All patient identifiers were removed from the examination data at the time data were transferred from each institution to the database.

The indications for CT colonography were varied and included the following: true screening studies, follow-up examinations for prior polyps, failed colonoscopy, and colonoscopy for patients who were at risk for colon polyps or carcinoma but were not deemed to be candidates for optical colonoscopy. Because of the multi-institutional nature of the study, the specific reasons for CT colonography were not known. Acquisition parameters for CT colonography varied across the sites (Table 1). The colons were generally fully cleansed, distended, and insufflated according to various protocols (Table 2). The image quality, however, varied because of the variability of acquisition parameters, presence of artifacts (breathing, hip implants, noise), and often considerable residual fecal material.


View this table:
[in this window]
[in a new window]

  		Table 1. Acquisition Characteristics for Multidetector CT Colonography at the Six CT Examination Sites

 

View this table:
[in this window]
[in a new window]

  		Table 2. Patient Preparation for CT Colonography

 
Polyps were characterized according to size as small (polyp <6 mm), medium (polyp ≥6 mm and ≤10 mm), and large (polyp >10 mm). Cases were identified as positive (at least one medium- or large-sized polyp) or negative (no polyps present). Optical colonoscopy was the reference standard for the majority of the cases (Table 3). For cases with incomplete colonoscopy or for which confirmation with optical colonoscopy was not available, two experienced readers who were not participating in the study independently identified the polyps before this study and served as the reference standard. Findings identified by only one reader were reviewed together; these were only the small polyps. If consensus could not be reached, these polyps were discarded since they are considered to have less clinical importance. Details for the CT colonography studies selected from the database and used for physician training and testing are shown in Table 3. The histologic type was not known for all polyps.


View this table:
[in this window]
[in a new window]

  		Table 3. CT Colonographic Studies Used for Physician Training and Testing

 
The CT colonographic studies were partitioned into two sets: a training set, to train the CAD system and for demonstration to and training of readers, and a test set. The examination data contributed from examination site B (n = 22) were assigned to the training set only and not to the test set because readers for the test set were from examination site B. The remaining CT colonographic studies (n = 238) were randomly partitioned into two equal groups of 119 each. The final training set consisted of 141 (22 + 119) CT studies; the testing set contained 119 studies.

In the training set, 67 of the 141 studies had positive results, with 106 polyps: 51 small polyps, 37 medium polyps, and 18 large polyps. From these, 23 CT colonographic studies were used for physician training (n = 20) and demonstration (n = 3); all 141 were used for CAD training. The CT colonographic studies used for physician training included 15 positive results, with 23 polyps (eight small, nine medium, and six large), and five negative results (Table 3). The CT colonographic studies in the training set were organized in increasing order of complexity both of three- and two-dimensional review and navigation—from having all the segments fully distended to poorly distended and collapsed. They were also organized according to lesion structure and position.

In the test set, 53 of the 119 CT studies had positive results, with 102 polyps: 52 small polyps, 32 medium polyps, and 18 large polyps. These studies were sequestered for testing of the CAD system and had not been seen by any of the seven readers. Given the investigative nature of our study and time constraints of the readers, 30 CT colonographic studies from 30 patients (mean age ± standard deviation, 62.5 years ± 9.98; age range, 44–80 years; 18 men and 12 women) were selected randomly from the 119 studies. Clinically, only two of these 30 studies were obtained at screening examinations. An enriched polyp population was desired; hence, 24 CT colonographic studies with positive findings were chosen, with 53 polyps (14 small, 23 medium, and 16 large) (Table 3). None of the CT studies with positive or negative results were excluded because of residual feces, fluid, motion or breathing artifacts, artifacts originating from implants, poor distention, or collapsed areas due to diverticulosis. Additionally, none of the readers had any knowledge of the origin of the images, the number of positive or negative CT colonographic results, or the number of polyps in each case. Finally, none of the readers were informed about the sensitivity or false-positive rate of CAD.

Study Protocol
The review of each CT colonographic study had an initial interpretation, during which the physicians evaluated the studies. Immediately after the initial interpretation, a second interpretation occurred in which the physicians reviewed the CAD findings. The readers were provided with data-recording booklets that contained a checklist and a means to record the reading time for the reader alone and for the additional time needed to review the CAD-proposed findings. Readers were instructed to ignore any polyp smaller than 5 mm. The case review was performed on a workstation by using Syngo Colonography (Siemens Medical Solution), with three dimensions as the primary setting and two dimensions for problem solving, or vice versa.

Initial interpretation.—The reader loaded both the supine and the prone images. Once these images were loaded, the reader recorded the time. The CAD program ran in the background while the reader assessed the CT study. Each finding was measured and located according to eight colonic segments (cecum, ascending, hepatic flexure, transverse, splenic flexure, descending, sigmoid, and rectum). A confidence score was assigned by using a five-point scale: score of 1, definitely not a polyp; 2, probably not a polyp; 3, indeterminate; 4, probably a polyp; and 5, definitely a polyp. Finally, the marks and annotations were saved and the time was recorded. The readers did not receive specific instructions for polyp measurement.

Second interpretation.—Immediately after the initial interpretation, the reader reviewed the candidates (ie, potential lesions) detected with CAD. If the CAD mark coincided with a reader's mark, the CAD mark was discarded but the finding was recorded. Any new CAD finding was reviewed in a manner similar to that used for those findings identified by the radiologist. After all the CAD markers were reviewed, the findings were saved and the time was recorded.

Reader Profile and Experience
The seven participating readers were from two institutions: reader site 1 (four readers), Temple University Hospital, Philadelphia, Pa, and reader site 2 (three readers), Cleveland Clinic Foundation, Cleveland, Ohio.

Reader site 1.—A senior attending abdominal radiologist (20 years of posttraining experience) had some experience with CT colonography using a primary two-dimensional method. This radiologist had participated as a reader in another CT colonographic investigation (50 examinations) but had no experience with clinical CT colonography or with the software used in this investigation. The other three readers—a senior attending abdominal radiologist (10 years of posttraining experience), a junior attending radiologist (1 year of posttraining experience), and a radiology resident in his 3rd year of training—had no experience with CT colonography experience prior to this study.

Reader site 2.—The senior attending abdominal radiologist, with 20 years of post-fellowship experience and some experience with CT colonography (n = 150 with colonoscopic comparison), had attended a course on CT colonography. The second radiologist, with 20 years of post-fellowship training, had limited CT colonography experience (10 cases) and had read studies in both two and three dimensions. This reader migrated to three-dimensional images as a primary reading mode by the time the test images were interpreted. The third reader, who had 12 years of post-fellowship training, had very limited CT colonography experience and used two-dimensional images as the primary reading mode.

Reader Demonstration and Training on CAD System
Reader site 1.—During the demonstration and training phases of the investigation, a dedicated application specialist (L.B.) reviewed the features and functionality of the system on three studies and helped the radiologist review subsequent studies. A total of 20 CT colonographic studies with known pathologic abnormalities were reviewed by each reader before the investigation began. At completion of the review of each study, the reader had access to a booklet that included pictures of the lesions in both transverse and coronal multiplanar reformations, in a rendered endoscopic view, and in a global view that showed a marker on the exact location of the lesion. Initially, all readers began reviewing by using the two-dimensional method but, after the first few studies, all switched to the three-dimensional method for the remainder of the studies and for the test phase.

Reader site 2.—The demonstration and training portion was compressed into a day for all three readers. As for site 1, readers at site 2 were shown the three demonstration CT colonographic studies and were subsequently allowed to review the 20 training CT colonographic studies. They had at their disposal similar booklets and feedback from the application specialist (L.B.). Two readers used the three-dimensional method as the primary form of interpretation for both the training and the test phases of the investigation; the third reader used the two-dimensional method as the primary form for both training and testing.

Test Phase
Reader site 1.—A total of 30 CT colonographic studies were reviewed in this phase. This phase was conducted over a period of 4 weeks; the readers were on a clinical service and were interrupted during the interpretations. Each radiologist reviewed two to five CT colonographic studies a day.

Reader site 2.—As with site 1, a total of 30 CT colonographic studies were reviewed in this phase. This phase was conducted in three consecutive sessions over 1 day for each reader, and 10 CT colonographic studies were read in each session (each session lasted approximately 4 hours, with a break taken during each session after 2 hours; an approximate 1-hour lunch break was taken). Two of the readers had no clinical responsibilities when the test images were interpreted; one of the readers was on a clinical service and was routinely interrupted during the interpretation sessions.

Statistical Analysis
We had intended to evaluate the data by using receiver operating characteristic analysis. However, the readers primarily used scores only at the ends of the confidence scale; thus, this analysis is not recommended (33). For each reader, the McNemar test, adjusted for clustered data, was used to compare the reader's sensitivity without and with CAD (34). The mean of the readers' sensitivities without and with CAD was compared (SAS, version 9.1; SAS Institute, Cary, NC) by using an analysis of variance approach for clustered data (http://www.bio.ri.ccf.org/html/rocanalysis.html, Cleveland, Ohio) (35). This method takes into account the variability and correlations between readers. The mean sensitivity and specificity can be interpreted as the accuracy of an average reader from the population. For purposes of analysis, a confidence score greater than 3 (a score of 3 was never used) was considered positive, and confidence scores of 1 and 2 were considered negative. We calculated the number of false-positive results per patient without and with CAD and compared them by using the Wilcoxon signed-rank test (SAS, version 9.1; SAS Institute). A P value of less than .05 was considered to indicate a statistically significant difference.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Improvement with CAD
The per-polyp sensitivity improved with CAD for all seven readers; the differences were statistically significant for three (reader 1: 0.69 without CAD, 0.85 with CAD, P = .006; reader 3: 0.77 without CAD, 0.90 with CAD, P = .038; and reader 5: 0.72 without CAD, 0.92 with CAD, P = .003) (Table 4). The mean sensitivity showed significant improvement (P = .0152) for polyp detection: from 0.810 without CAD (95% confidence interval [CI]: 0.685, 0.934) (Table 4) to 0.908 with CAD (95% CI: 0.806, 1.00) (Table 4). The 95% CI for the improvement in sensitivity due to CAD was 0.027–0.171.


View this table:
[in this window]
[in a new window]

  		Table 4. Reader Sensitivities without and with CAD

 
Without CAD, the number of false-positive results per patient ranged from 0.40 to 1.17 (mean, 0.70 per patient) (Table 5). With CAD, the number of false-positive results per patient ranged from 0.53 to 1.73 (mean, 0.96) (Table 5). The 95% CI for the average increase in false-positive results per patient due to CAD was –0.39 to 0.91. For three of the seven readers, the number of false-positive results was significantly higher with CAD than without CAD (Table 5).


View this table:
[in this window]
[in a new window]

  		Table 5. False-Positive Findings for Readers without and with CAD

 
Detection by Polyp Size
The detection of each of the polyps with CAD and cumulative detection by the readers in general showed that both the readers and CAD identified most medium-size polyps (Fig 2). Both the readers and the CAD program had some difficulty identifying smaller polyps, whereas CAD was less successful in depicting lesions larger than 25 mm in size. Readers rejected true-positive CAD candidates (Fig 3), and both readers and CAD missed polyps (Fig 4). CAD performance was 0.87 for medium-size polyps, 0.82 for large polyps, and 0.85 overall for polyps between 6 mm and 25 mm, with a false-positive rate of 6.93 per patient presented to the reader.


Figure 2
View larger version (24K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2: Individual polyp detection by readers alone ({blacksquare}) and with CAD ({circ}), as well as polyps missed by reader or CAD. The frequency shows how often a given polyp of a specific size was detected by readers and with CAD. For instance, on the x-axis, looking at one 6-mm polyp (there were three polyps of this size), we can see that it was detected with CAD (C Found) and by all seven readers; a 7.0-mm polyp was missed with CAD (C Missed) but was detected by all readers. Finally, all polyps but one in the range 6.0–9.3 mm were detected with CAD; some were detected by as many as seven or as few as four readers. R Missed = individual polyp missed by a reader.

 

Figure 3
View larger version (88K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3: CT colonographic images show transverse two-dimensional (left) and endoluminal three-dimensional (right) views of a 7.4-mm polyp in the transverse colon (2b marks polyp identified by one reader). The polyp was found with CAD but was rejected by two readers.

 

Figure 4
View larger version (84K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4: CT colonographic images show transverse two-dimensional (left) and endoluminal three-dimensional (right) views of a 9.8-mm polyp in the cecum (3b marks polyp identified by one reader). The polyp was missed or dismissed by both readers and CAD.

 
Review Time
The time to review the CT colonographic studies without CAD varied from a mean of slightly more than 10 minutes to nearly 20 minutes (Fig 5), with a mean time for all readers of slightly more than 14 minutes. The mean total time to review the CT colonographic studies for each reader before and after CAD varied from almost 13 minutes to more than 23 minutes (Fig 5). The mean total time for all the readers to review the CT studies without and with CAD was approximately 18 minutes; the mean time needed to review each CAD candidate was approximately 23 seconds.


Figure 5
View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5: Graph shows mean reader time plus mean time to review CAD findings. Ave = mean time for the seven readers.

 
The mean number of CAD candidates varied among the three examination groups: group 1, 16 candidates (range, 5–39 candidates); group 2, 9.6 candidates (range, 4–19 candidates); and group 3, 6.2 candidates (range, 3–12 candidates). The mean time to review all the CAD candidates also varied among the three examination groups: group 1, 5.4 minutes (range, 2–10 minutes); group 2, 3.4 minutes (range, 2–5 minutes); and group 3, 2.5 minutes (range, 1–5 minutes). The mean time to review each CAD candidate was very similar in the three examination groups: group 1, 22.7 seconds (range, 15.4–30.0 seconds); group 2, 23.3 seconds (range, 15.0–33.3 seconds); and group 3, 23.6 seconds (range, 15.0–36.0 seconds).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Double readings have been shown to improve polyp detection with air-contrast barium enema (36,37) and cancer detection with mammography (38); however, from a practical perspective, in a busy practice, double reading may not be a viable alternative. CAD technology proposes to fulfill such a second-reader role and, in fact, is used routinely in many breast centers (24). Destounis et al (38) have shown that CAD can even improve cancer detection over double-read mammograms. Hence, as demonstrated in several preliminary CT colonographic studies (29,39–47), our study also supports the evidence of the use of CAD as a second reader. Furthermore, given the level of experience of the participating readers in our study, it appears that CAD may be extremely valuable for the general radiologist.

In our investigation, CAD depicted most polyps between 6 and 25 mm in size, but CAD did not depict lesions larger than 25 mm with regularity. This performance is consistent with that reported in previous studies (29,31,32). Further, the system was not designed to depict such large polyps; these large lesions were detected by most of the radiologists. Current developments of this CAD system are aimed at improving detection in polyps larger than 25 mm and up to 40 mm. Most radiologists should be able to detect large polyps and cancers. Further, we believe that the main role of CAD is to assist in the detection of a medium-size polyp.

Even though CAD demonstrated high sensitivity for the polyps between 6 and 25 mm, it also resulted in many more false-positive findings. Thus, we do not believe that it can replace a primary radiologist's interpretation of CT colonographic results. Conversely, we believe that the additional polyps depicted with CAD, especially for inexperienced readers, justifies the effort in reviewing the findings and the additional false-positive results. Newer versions of the software have reduced the number of false-positive marks (31,32).

The time taken to assess the CAD candidates does not appear to be excessive. The overall mean time of nearly 3.5 minutes to review the CAD markings involved only 19% of the total time (mean total time, 18 minutes 54 seconds) taken to assess the CT colonographic studies without and with CAD. Although we did not specifically assess the issue, many of the CAD candidates were spurious marks outside the colon and could easily be eliminated. The mean time to assess CAD candidates was approximately 23 seconds for all three groups.

There were limitations to our study. The study population was small, but we were able to enhance our ability to detect differences by using multiple readers. The population was polyp enriched, but enrichment of a study population is an accepted practice to test any system (48), especially with CAD (enriched populations were used to test CAD in mammography). The time period allowed for reading at the two reader sites differed: at site 1, the readers had 4 weeks in which to read 30 studies, whereas at site 2, the reading took place in only 1 day. All the readers at site 1 and one reader at site 2 were routinely interrupted when interpreting the test images. Although it would have been ideal to have a uniform, uninterrupted time period to interpret the images, we believe that the experience in our study reflects the reality of a day-to-day busy practice. Most of the readers were inexperienced; only one had some experience in clinical cases.

One could argue that any help would have improved the reader sensitivity to detect polyps; however, we would contend that our reader profile reflects the general radiology community. In a recent study of more than 1000 screening CT colonographic examinations, CAD was equivalent to optical colonoscopy for adenomas larger than 8 mm (46); another reported that CAD helped improve the performance of experienced CT colonographic readers (47). In our investigation, CAD also helped improve polyp detection for even the most experienced radiologist (although that improvement was not statistically significant). Our study is also limited in that we do not know the indication for the CT colonography or the histologic type of the polyps used for CAD training or testing.

In conclusion, results of our preliminary study suggest that CAD for CT colonography significantly improves the per-polyp detection for less-experienced readers. Whereas CAD increases the number of false-positive results per patient, the potential increase in sensitivity and the relatively short time spent reviewing the CAD findings appear to justify the use of CAD.


   ADVANCE IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE
 References
 


   IMPLICATION FOR PATIENT CARE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
References
 


   FOOTNOTES
 

Abbreviations: CAD = computed-aided detection • CI = confidence interval

Guarantor of integrity of entire study, M.E.B.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, M.E.B., L.B., A.J.; clinical studies, M.E.B., C.D., R.M.K., E.M.R., D.M.E., A.B., D.F.C., M.M.; experimental studies, M.E.B., L.B., P.C., A.J., S.L., M.M.; statistical analysis, N.A.O., A.J.; and manuscript editing, M.E.B., L.B., N.A.O., E.M.R., D.M.E., P.C., A.J., S.L., D.F.C., M.M.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
ADVANCE IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 

  1. Vining DJ, Gelfand DW. Noninvasive colonoscopy using helical CT scanning, 3D reconstruction, and virtual reality. Presented at the 23rd annual meeting and postgraduate course of the Society of Gastrointestinal Radiologists, Maui, Hawaii, February 13–18, 1994.
  2. Hara AK, Johnson CD, Reed JE, Ehman RL, Ilstrup DM. Colorectal polyp detection with CT colonography: two- versus three-dimensional techniques. Work in progress. Radiology 1996;200(1):49–54.
  3. Johnson CD, Dachman AH. CT colonography: the next colon screening examination? Radiology 2000;216(2):331–341. [Abstract/Free Full Text]
  4. McFarland EG, Pilgram TK, Brink JA, et al. CT colonography: multiobserver diagnostic performance. Radiology 2002;225(2):380–390. [Abstract/Free Full Text]
  5. Sosna J, Morrin MM, Kruskal JB, Lavin PT, Rosen MP, Raptopoulos V. CT colonography of colorectal polyps: a metaanalysis. AJR Am J Roentgenol 2003;181(6):1593–1598. [Abstract/Free Full Text]
  6. Pickhardt PJ, Choi JR, Hwang I, et al. Computed tomographic virtual colonoscopy to screen for colorectal neoplasia in asymptomatic adults. N Engl J Med 2003;349(23):2191–2200. [Abstract/Free Full Text]
  7. Cotton PB, Durkalski VL, Pineau BC, et al. Computed tomographic colonography (virtual colonoscopy): a multicenter comparison with standard colonoscopy for detection of colorectal neoplasia. JAMA 2004;291(14):1713–1719. [Abstract/Free Full Text]
  8. Edwards JT, Mendelson RM, Fritschi L, et al. Colorectal neoplasia screening with CT colonography in average-risk asymptomatic subjects: community-based study. Radiology 2004;230(2):459–464. [Abstract/Free Full Text]
  9. Macari M, Bini EJ, Jacobs SL, et al. Colorectal polyps and cancers in asymptomatic average-risk patients: evaluation with CT colonography. Radiology 2004;230(3):629–636. [Abstract/Free Full Text]
  10. Macari M, Bini EJ, Jacobs SL, et al. Significance of missed polyps at CT colonography. AJR Am J Roentgenol 2004;183(1):127–134. [Abstract/Free Full Text]
  11. Scott RG, Edwards JT, Fritschi L, Foster NM, Mendelson RM, Forbes GM. Community-based screening by colonoscopy or computed tomographic colonography in asymptomatic average-risk subjects. Am J Gastroenterol 2004;99(6):1145–1151. [CrossRef][Medline]
  12. Rockey DC, Paulson E, Niedzwiecki D, et al. Analysis of air contrast barium enema, computed tomographic colonography, and colonoscopy: prospective comparison. Lancet 2005;365(9456):305–311. [Medline]
  13. Mulhall BP, Veerappan GR, Jackson JL. Meta-analysis: computed tomographic colonography. Ann Intern Med 2005;142(8):635–650. [Abstract/Free Full Text]
  14. Pickhardt PJ, Taylor AJ, Johnson GL, et al. Building a CT colonography program: necessary ingredients for reimbursement and clinical success. Radiology 2005;235(1):17–20. [Free Full Text]
  15. Imperiale TF. Can computed tomographic colonography become a "good" screening test? Ann Intern Med 2005;142(8):669–670. [Free Full Text]
  16. Hardacre JM, Ponsky JL, Baker ME. Colonoscopy vs CT colonography to screen for colorectal neoplasia in average-risk patients. Surg Endosc 2005;19(3):448–456. [CrossRef][Medline]
  17. Soto JA, Barish MA, Yee J. Reader training in CT colonography: how much is enough? Radiology 2005;237(1):26–27. [Free Full Text]
  18. Taylor SA, Halligan S, Burling D, et al. CT colonography: effect of experience and training on reader performance. Eur Radiol 2004;14(6):1025–1033. [CrossRef][Medline]
  19. Fenlon HM. CT colonography: pitfalls and interpretation. Abdom Imaging 2002;27(3):284–291. [Medline]
  20. Fidler JL, Fletcher JG, Johnson CD. Understanding interpretive errors in radiologists learning computed tomography colonography. Acad Radiol 2004;11:750–756. [CrossRef][Medline]
  21. Petrick N, Sahiner B, Chan HP, Helvie MA, Paquerault S, Hadjiiski LM. Breast cancer detection: evaluation of a mass-detection algorithm for computer-aided diagnosis—experience in 263 patients. Radiology 2002;224(1):217–224. [Abstract/Free Full Text]
  22. Helvie MA, Hadjiiski L, Makariou E, et al. Sensitivity of noncommercial computer-aided detection system for mammographic breast cancer detection: pilot clinical trial. Radiology 2004;231(1):208–214. [Abstract/Free Full Text]
  23. Baker JA, Rosen EL, Lo JY, Gimenez EI, Walsh R, Soo MS. Computer-aided detection (CAD) in screening mammography: sensitivity of commercial CAD systems for detecting architectural distortion. AJR Am J Roentgenol 2003;181(4):1083–1088. [Abstract/Free Full Text]
  24. Cupples TE, Cunningham JE, Reynolds JC. Impact of computer-aided detection in a regional screening mammography program. AJR Am J Roentgenol 2005;185(4):944–950. [Abstract/Free Full Text]
  25. Awai K, Murao K, Ozawa A, et al. Pulmonary nodules at chest CT: effect of computer-aided diagnosis on radiologists' detection performance. Radiology 2004;230(2):347–352. [Abstract/Free Full Text]
  26. Rubin GD, Lyo JK, Paik DS, et al. Pulmonary nodules on multi-detector row CT scans: performance comparison of radiologists and computer-aided detection. Radiology 2005;234(1):274–283. [Abstract/Free Full Text]
  27. Bae KT, Kim JS, Na YH, Kim KG, Kim JH. Pulmonary nodules: automated detection on CT images with morphologic matching algorithm—preliminary results. Radiology 2005;236(1):286–293. [Abstract/Free Full Text]
  28. Li F, Arimura H, Suzuki K, et al. Computer-aided detection of peripheral lung cancers missed at CT: ROC analyses without and with localization. Radiology 2005;237(2):684–690. [Abstract/Free Full Text]
  29. Bogoni L, Cathier P, Dundar M, et al. Computer-aided detection (CAD) for CT colonography: a tool to address a growing need. Br J Radiol 2005;78(spec no 1):S57–S62. [Abstract/Free Full Text]
  30. Dundar M, Fung G, Bogoni L, Macari M, Megibow A, Rao B. A methodology for training and validating a CAD system and potential pitfalls. CARS July 2004; 1010–1014.
  31. Graser A, Bogoni L, Becker CR, Reiser MF. Computer-aided detection (CAD) in MDCT colonography: evaluation of the performance of a prototype system in more than 100 cases [abstr]. In: Radiological Society of North America Scientific Assembly and Annual Meeting Program. Oak Brook, Ill: Radiological Society of North America, 2005; 337.
  32. Mang T, Peloschek P, Plank C, et al. Effect of computer aided detection as a second reader in multidetector CT colonography: a multiobserver study. Presented at the 2006;European Congress of Radiology, Vienna, Austria, March 3–7, 2006.
  33. Zhou XH, Obuchowski NA, McClish DK. Statistical methods in diagnostic medicine. New York, NY: Wiley, 2002;133–136.
  34. Obuchowski NA. On the comparison of correlated proportions for clustered data. Stat Med 1998;17(13):1495–1507. [CrossRef][Medline]
  35. Obuchowski NA. Multireader, multimodality receiver operating characteristic curve studies: hypothesis testing and sample size estimation using an analysis of variance approach with dependent observations. Acad Radiol 1995;2(suppl 1):S22–S29. [Medline]
  36. Markus JB, Somers S, O'Malley BP, Stevenson GW. Double-contrast barium enema studies: effect of multiple reading on perception error. Radiology 1990;175(1):155–156. [Abstract/Free Full Text]
  37. Johnson CD, MacCarty RL, Welch TJ, et al. Comparison of the relative sensitivity of CT colonography and double-contrast barium enema for screen detection of colorectal polyps. Clin Gastroenterol Hepatol 2004;2(4):314–321. [CrossRef][Medline]
  38. Destounis SV, DiNitto P, Logan-Young W, Bonaccio E, Zuley ML, Willison KM. Can computer-aided detection with double reading of screening mammograms help decrease the false-negative rate? Initial experience. Radiology 2004;232(2):578–584.
  39. Summers RM, Jerebko AK, Franaszek M, Malley JD, Johnson CD. Colonic polyps: complementary role of computer-aided detection in CT colonography. Radiology 2002;225(2):391–399. [Abstract/Free Full Text]
  40. Summers RM. Challenges for computer-aided diagnosis for CT colonography. Abdom Imaging 2002;27(3):268–274. [Medline]
  41. Yoshida H, Masutani Y, MacEneaney P, Rubin DT, Dachman AH. Computerized detection of colonic polyps at CT colonography on the basis of volumetric features: pilot study. Radiology 2002;222(2):327–336. [Abstract/Free Full Text]
  42. Yoshida H, Nappi J, MacEneaney P, Rubin DT, Dachman AH. Computer-aided diagnosis scheme for detection of polyps at CT colonography. RadioGraphics 2002;22(4):963–979. [Abstract/Free Full Text]
  43. Mani A, Napel S, Paik DS, et al. Computed tomography colonography: feasibility of computer-aided polyp detection in a "first-reader" paradigm. J Comput Assist Tomogr 2004;28(3):318–326. [CrossRef][Medline]
  44. Yoshida H, Dachman AH. CAD techniques, challenges, and controversies in computed tomographic colonography. Abdom Imaging 2005;30(1):26–41. [CrossRef][Medline]
  45. Summers RM, Franaszek M, Miller MT, Pickhardt PJ, Choi JR, Schindler WR. Computer-aided detection of polyps on oral contrast-enhanced CT colonography. AJR Am J Roentgenol 2005;184(1):105–108. [Free Full Text]
  46. Summers RM, Yao J, Pickhardt PJ, et al. Computed tomographic virtual colonoscopy computer-aided polyp detection in a screening population. Gastroenterology 2005;129:1832–1844. [CrossRef][Medline]
  47. Taylor SA, Halligan S, Burling D, et al. Computer-assisted reader software versus expert reviewers for polyp detection on CT colonography. AJR Am J Roentgenol 2006;186:696–702. [Abstract/Free Full Text]
  48. Zhou XH, Obuchowski NA, McClish DK. Statistical methods in diagnostic medicine. New York, NY: Wiley, 2002; 63–66.



This article has been cited by other articles:


Home page
 Am. J. Roentgenol.Home page
S. A. Taylor, J. Brittenden, J. Lenton, H. Lambie, A. Goldstone, P. N. Wylie, D. Tolan, D. Burling, L. Honeyfield, P. Bassett, et al.
Influence of Computer-Aided Detection False-Positives on Reader Performance and Diagnostic Confidence for CT Colonography
Am. J. Roentgenol., June 1, 2009; 192(6): 1682 - 1689.
[Abstract] [Full Text] [PDF]

Home page
 JNMHome page
M. Sadik, I. Hamadeh, P. Nordblom, M. Suurkula, P. Hoglund, M. Ohlsson, and L. Edenbrandt
Computer-Assisted Interpretation of Planar Whole-Body Bone Scans
J. Nucl. Med., December 1, 2008; 49(12): 1958 - 1965.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Roentgenol.Home page
R. M. Summers, L. R. Handwerker, P. J. Pickhardt, R. L. Van Uitert, K. K. Deshpande, S. Yeshwant, J. Yao, and M. Franaszek
Performance of a Previously Validated CT Colonography Computer-Aided Detection System in a New Patient Population
Am. J. Roentgenol., July 1, 2008; 191(1): 168 - 174.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Baker, M. E.
Right arrow Articles by Macari, M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Baker, M. E.
Right arrow Articles by Macari, M.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE