Suryakanth R. Gurudu

A Meta-Analysis of the Yield of Capsule Endoscopy Compared to Other Diagnostic Modalities in Patients with Non-Stricturing Small Bowel Crohn's Disease

OBJECTIVES:Capsule endoscopy (CE) allows for direct evaluation of the small bowel mucosa in patients with Crohns disease (CD). A number of studies have revealed significantly improved yield for CE over other modalities for the diagnosis of CD, but as sample sizes have been small, the true degree of benefit is uncertain. Additionally, it is not clear whether patients with a suspected initial presentation of CD and those with suspected recurrent disease are equally likely to benefit from CE. The aim of this study was to evaluate the yield of CE compared with other modalities in symptomatic patients with suspected or established CD using meta-analysis.METHODS:We performed a recursive literature search of prospective studies comparing the yield of CE to other modalities in patients with suspected or established CD. Data on yield among various modalities were extracted, pooled, and analyzed. Incremental yield (IY) (yield of CE − yield of comparative modality) and 95% confidence intervals (95% CI) of CE over comparative modalities were calculated. Subanalyses of patients with a suspected initial presentation of CD and those with suspected recurrent disease were also performed.RESULTS:Nine studies (n = 250) compared the yield of CE with small bowel barium radiography for the diagnosis of CD. The yield for CE versus barium radiography for all patients was 63% and 23%, respectively (IY = 40%, p < 0.001, 95% CI = 28–51%). Four trials compared the yield of CE to colonoscopy with ileoscopy (n = 114). The yield for CE versus ileoscopy for all patients was 61% and 46%, respectively (IY = 15%, p= 0.02, 95% CI = 2–27%). Three studies compared the yield of CE to computed tomography (CT) enterography/CT enteroclysis (n = 93). The yield for CE versus CT for all patients was 69% and 30%, respectively (IY = 38%, p= 0.001, 95% CI = 15–60%). Two trials compared CE to push enteroscopy (IY = 38%, p < 0.001, 95% CI = 26–50%) and one trial compared CE to small bowel magnetic resonance imaging (MRI) (IY = 22%, p= 0.16, 95% CI =−9% to 53%). Subanalysis of patients with a suspected initial presentation of CD showed no statistically significant difference between the yield of CE and barium radiography (p= 0.09), colonoscopy with ileoscopy (p= 0.48), CT enterography (p= 0.07), or push enteroscopy (p= 0.51). Subanalysis of patients with established CD with suspected small bowel recurrence revealed a statistically significant difference in yield in favor of CE compared with all other modalities (barium radiography (p < 0.001), colonoscopy with ileoscopy (p= 0.002), CT enterography (p < 0.001), and push enteroscopy (p < 0.001)).CONCLUSIONS:In study populations, CE is superior to all other modalities for diagnosing non-stricturing small bowel CD, with a number needed to test (NNT) of 3 to yield one additional diagnosis of CD over small bowel barium radiography and NNT = 7 over colonoscopy with ileoscopy. These results are due to a highly significant IY with CE over all other modalities in patients with established non-stricturing CD being evaluated for a small bowel recurrence. While there was no significant difference seen between CE and alternate modalities for diagnosing small bowel CD in patients with a suspected initial presentation of CD, the trend toward significance for a number of modalities suggests the possibility of a type II error. Larger studies are needed to better establish the role of CE for diagnosing small bowel CD in patients with a suspected initial presentation of CD.

Capsule endoscopy has a significantly higher diagnostic yield in patients with suspected and established small-bowel Crohn's disease: a meta-analysis

OBJECTIVES:Capsule endoscopy (CE) has demonstrated superior performance compared with other modalities in its ability to detect early small-bowel (SB) Crohns disease (CD), especially when ileoscopy is negative or unsuccessful. The aim of this study was to evaluate the diagnostic yield of CE compared with other modalities in patients with suspected and established CD using a meta-analysis.METHODS:A thorough literature search for prospective studies comparing the diagnostic yield of CE with other modalities in patients with CD was undertaken. Other modalities included push enteroscopy (PE), colonoscopy with ileoscopy (C+IL), SB radiography (SBR), computed tomography enterography (CTE), and magnetic resonance enterography (MRE). Data on diagnostic yield among various modalities were extracted, pooled, and analyzed. Data on patients with suspected and established CD were analyzed separately. Weighted incremental yield (IYW) (diagnostic yield of CE−diagnostic yield of comparative modality) and 95% confidence intervals (CIs) of CE over comparative modalities were calculated.RESULTS:A total of 12 trials (n=428) compared the yield of CE with SBR in patients with CD. Eight trials (n=236) compared CE with C+IL, four trials (n=119) compared CE with CTE, two trials (n=102) compared CE with PE, and four trials (n=123) compared CE with MRE. For the suspected CD subgroup, several comparisons met statistical significance. Yields in this subgroup were CE vs. SBR: 52 vs. 16% (IYw=32%, P<0.0001, 95% CI=16–48%), CE vs. CTE: 68 vs. 21% (IYw=47%, P<0.00001, 95% CI=31–63%), and CE vs. C+IL: 47 vs. 25% (IYw=22%, P=0.009, 95% CI=5–39%). Statistically significant yields for CE vs. an alternate diagnostic modality in established CD patients were seen in CE vs. PE: 66 vs. 9% (IYw=57%, P<0.00001, 95% CI=43–71%), CE vs. SBR: 71 vs. 36% (IYw=38%, P<0.00001, 95% CI=22–54%), and in CE vs. CTE: 71 vs. 39% (IYw=32%, P=<0.0001, 95% CI=16–47%).CONCLUSIONS:Our meta-analysis demonstrates that CE is superior to SBR, CTE, and C+IL in the evaluation of suspected CD patients. CE is also a more effective diagnostic tool in established CD patients compared with SBR, CTE, and PE.

Convolutional Neural Networks for Medical Image Analysis: Full Training or Fine Tuning?

Training a deep convolutional neural network (CNN) from scratch is difficult because it requires a large amount of labeled training data and a great deal of expertise to ensure proper convergence. A promising alternative is to fine-tune a CNN that has been pre-trained using, for instance, a large set of labeled natural images. However, the substantial differences between natural and medical images may advise against such knowledge transfer. In this paper, we seek to answer the following central question in the context of medical image analysis: Can the use of pre-trained deep CNNs with sufficient fine-tuning eliminate the need for training a deep CNN from scratch? To address this question, we considered four distinct medical imaging applications in three specialties (radiology, cardiology, and gastroenterology) involving classification, detection, and segmentation from three different imaging modalities, and investigated how the performance of deep CNNs trained from scratch compared with the pre-trained CNNs fine-tuned in a layer-wise manner. Our experiments consistently demonstrated that 1) the use of a pre-trained CNN with adequate fine-tuning outperformed or, in the worst case, performed as well as a CNN trained from scratch; 2) fine-tuned CNNs were more robust to the size of training sets than CNNs trained from scratch; 3) neither shallow tuning nor deep tuning was the optimal choice for a particular application; and 4) our layer-wise fine-tuning scheme could offer a practical way to reach the best performance for the application at hand based on the amount of available data.Training a deep convolutional neural network (CNN) from scratch is difficult because it requires a large amount of labeled training data and a great deal of expertise to ensure proper convergence. A promising alternative is to fine-tune a CNN that has been pre-trained using, for instance, a large set of labeled natural images. However, the substantial differences between natural and medical images may advise against such knowledge transfer. In this paper, we seek to answer the following central question in the context of medical image analysis: Can the use of pre-trained deep CNNs with sufficient fine-tuning eliminate the need for training a deep CNN from scratch? To address this question, we considered four distinct medical imaging applications in three specialties (radiology, cardiology, and gastroenterology) involving classification, detection, and segmentation from three different imaging modalities, and investigated how the performance of deep CNNs trained from scratch compared with the pre-trained CNNs fine-tuned in a layer-wise manner. Our experiments consistently demonstrated that 1) the use of a pre-trained CNN with adequate fine-tuning outperformed or, in the worst case, performed as well as a CNN trained from scratch; 2) fine-tuned CNNs were more robust to the size of training sets than CNNs trained from scratch; 3) neither shallow tuning nor deep tuning was the optimal choice for a particular application; and 4) our layer-wise fine-tuning scheme could offer a practical way to reach the best performance for the application at hand based on the amount of available data.

Comparison of the yield and miss rate of narrow band imaging and white light endoscopy in patients undergoing screening or surveillance colonoscopy: a meta-analysis.

OBJECTIVES:Colonoscopy has an appreciable miss rate for adenomas and colorectal cancer. The goal of advanced endoscopic imaging is to improve lesion detection. Compared with standard definition, high-definition (HD) colonoscopes have the advantage of increased field of visualization and higher resolution; narrow band imaging (NBI) utilizes narrow band filters for enhanced visualization of surface architecture and capillary pattern. The objective of this study was to compare the yield and miss rates of HD-NBI and HD-WLE (white light endoscopy) for the detection of colon polyps using meta-analysis.METHODS:A recursive literature search of randomized controlled trials (RCTs) comparing the yield of HD-NBI and HD-WLE for detection of colon polyps in patients undergoing screening/surveillance colonoscopy. Authors were contacted for missing data. In RCT with tandem colonoscopy (RCT-t), findings from the first-pass examinations were used in the yield analysis and from the tandem pass for the miss rate analysis. Data on the yield of polyps were extracted, pooled, and analyzed using RevMan 4.2.9 software. Odds ratio (OR) and 95% confidence intervals (CIs) for the pooled data for the yield and miss rates of NBI and WLE were calculated. A fixed effect model (FEM) was used for analyses without, and a random effect model (REM) for analyses with heterogeneity.RESULTS:The yield analysis revealed no significant difference between HD-NBI and HD-WLE for the detection of adenomas (six studies; n=2,284; OR: 1.01; CI: 0.74–1.37; REM); patients with polyps (six studies; n=2,275; OR: 1.15; CI: 0.8–1.64; REM); patients with adenomas (four studies; n=2,177; OR: 1.0; CI: 0.83–1.20; FEM); detection of adenomas <10 mm (five studies; n=1,618; OR: 1.32; CI: 0.92–1.88; FEM); flat adenomas (five studies; n=1,675; OR: 1.26; CI: 0.62–2.57; REM); and flat adenomas per patient (five studies; n=2,200; OR: 1.63; CI: 0.71–3.74; REM). The miss rate analysis revealed no difference in polyp miss rate (three studies; n=524; OR: 1.17; CI: 0.8–1.71; FEM) or adenoma miss rate (three studies; n=524; OR: 0.65; CI: 0.4–1.06; FEM) between the two techniques.CONCLUSIONS:Compared with HD-WLE, HD-NBI does not increase the yield of colon polyps, adenomas, or flat adenomas, nor does it decrease the miss rate of colon polyps or adenomas in patients undergoing screening/surveillance colonoscopy.

Increased adenoma detection rate with system-wide implementation of a split-dose preparation for colonoscopy

BACKGROUND Recent studies using split-dose preparations (SDPs) suggest a significant improvement in the quality of preparation and patient compliance. However, the effects of SDP on other quality indicators of colonoscopy, such as cecal intubation and adenoma detection rates, have not been previously reported, to our knowledge. OBJECTIVE The primary objective of this study was to compare polyp detection rates (PDRs) and adenoma detection rates (ADRs) before and after the implementation of an SDP as the preferred bowel preparation. The secondary objectives were to compare the quality of the preparation and colonoscopy completion rates before and after implementation of the SDP. DESIGN Retrospective study. SETTING Tertiary care medical center. PATIENTS Patients undergoing colonoscopy for screening and surveillance of colon polyps and cancer. INTERVENTIONS System-wide implementation of SDP. RESULTS A total of 3560 patients in the pre-SDP group and 1615 patients in the post-SDP group were included in the study. SDP use increased significantly from 9% to 74% after implementation. In comparison with the pre-SDP group, both PDRs (44.1%-49.5%; P < .001) and ADRs (26.7%-31.8%; P < .001) significantly improved in the post-SDP group. The colonoscopy completion rate significantly increased from 93.6% to 95.5% in the post-SDP group (P = .008). Bowel preparation quality also improved significantly (P < .001) in the post-SDP group. LIMITATIONS Retrospective design; not all endoscopists were the same in both periods. CONCLUSIONS System-wide implementation of an SDP as the primary choice for colonoscopy significantly improved both PDRs and ADRs, overall quality of the preparation, and colonoscopy completion rates.

Treatment of gastric antral vascular ectasia (watermelon stomach) with endoscopic band ligation

BACKGROUND Gastric antral vascular ectasia (GAVE) is characterized by mucosal and submucosal vascular ectasia causing recurrent GI hemorrhage. Treatment of GAVE with endoscopic thermal therapy (ETT) requires multiple sessions for destruction of vascular ectasia and control of bleeding. Endoscopic band ligation (EBL) has become the standard treatment of varices because it effectively obliterates the submucosal plexus of esophageal varices with an acceptably low rate of complications. Additionally, EBL has been used for control of bleeding from other GI vascular lesions. In patients with GAVE and recurrent GI hemorrhage, EBL may offer an alternative to ETT for treatment of large areas of diseased mucosa and submucosa. OBJECTIVE Our purpose was to compare EBL (n = 9) with ETT (n = 13) for the treatment of bleeding from GAVE. DESIGN Observational comparative study. PATIENTS Patients with gastric antral vascular ectasia with occult or overt bleeding. SETTING Mayo Clinic Arizona, a multispecialty academic medical center. INTERVENTION EBL or ETT with argon plasma coagulation or electrocautery. MAIN OUTCOME AND MEASUREMENTS Number of treatments to cessation of bleeding and posttreatment hemoglobin, hospitalization, and transfusion requirement. RESULTS There were no significant differences in the demographics, clinical presentation, associated portal hypertension, or mean hemoglobin values or the mean number of transfusions or hospitalizations between the 2 groups before treatment. Four patients in the EBL group had failed prior ETT. Compared with ETT, in exploratory statistical testing EBL had a significantly higher rate of bleeding cessation (67% vs 23%, P = .04), fewer treatment sessions required for cessation of bleeding (1.9 vs 4.7, P = .05), a greater increase in hemoglobin values (2.8 g/dL vs 0.9 g/dL, P = .05), a greater decrease in transfusion requirements (-12.7 vs -5.2, P = .02), and a greater decrease in hospital admissions (-2.6 vs -0.5, P = .02) during the follow-up period. Analysis of covariance showed significantly superior efficacy of EBL for cessation of bleeding, postprocedure transfusion, and hospitalization. One patient in the EBL group had postprocedure emesis and 1 in the ETT group had immediate post procedure bleeding. All patients in the EBL group had complete mucosal healing with minimal residual GAVE at follow-up endoscopy failed post-EBL. CONCLUSIONS Our initial experience suggests that EBL is superior to ETT for the management of GAVE. EBL required fewer treatment sessions for control of bleeding, had higher rates for cessation of bleeding, had a reduction in hospitalizations and transfusion requirements, and allowed for a significant increase in hemoglobin values.

Using CT Enterography to Monitor Crohn's Disease Activity: A Preliminary Study

OBJECTIVE The purpose of our study was to determine whether imaging changes of Crohns disease at sequential CT enterography examinations correlate with disease progression or regression. MATERIALS AND METHODS Forty CT enterography examinations in 20 patients (12 women, eight men; mean age, 55.5 years) with known Crohns disease were retrospectively evaluated by a radiologist who was blinded to the clinical history. One radiologist determined whether imaging findings of Crohns disease were present and, if so, whether the findings progressed, regressed, or remained stable between examinations. CT enterography findings were then compared with disease progression or regression based on symptoms and clinical follow-up. Direct comparison of CT enterography and endoscopy was also performed when available. RESULTS Disease progression or regression by CT enterography correlated with symptoms in 16 of 20 (80%) patients. Specifically, CT enterography and symptoms agreed in 12 patients with clinical disease progression, two patients with clinical regression, and two with clinically stable disease. In four of 20 (20%) patients, symptoms progressed although CT enterography findings were negative (n = 2) or improved (n = 2). No treatment change was initiated; and at follow-up, three of four patients were improved and the remaining patient was stable symptomatically. Sixteen ileoscopies were attempted in 12 patients; however, four examinations did not reach the ileum. In the remaining examinations, endoscopy correlated with CT enterography in all cases (12/12, 100%) and with symptoms in nine of 12 (75%) cases. The weighted kappa statistic, which measures the chance-adjusted agreement between CT enterography and symptoms, was 0.57 (95% CI, 0.20-0.94). CONCLUSION This preliminary study indicates that imaging changes between CT enterography examinations have excellent potential for reliably monitoring Crohns disease progression or regression.

Polyp and Adenoma Detection Rates in the Proximal and Distal Colon

OBJECTIVES:Little is known about the correlation between the polyp detection rate (PDR) and the adenoma detection rate (ADR) in individual colonic segments. The adenoma-to-polyp detection rate quotient (APDRQ) has been utilized in retrospective study as a constant to estimate ADR from PDR. It has been previously stated that diminutive polyps in the rectum are more likely to be non-adenomatous, compared with more proximal segments, yet the APDRQ uses data from the entire colon. We sought to characterize and compare ADR and PDR in each colonic segment, estimate ADR using the conversion factor, APDRQ, and assess the correlation between estimated and actual ADR for each colonic segment.METHODS:As part of a quality improvement program, a retrospective chart review was conducted of all outpatient colonoscopies performed by 20 gastroenterologists between 1 October 2010 and 31 March 2011 at a single academic tertiary-care referral center. PDR, ADR, and the APDRQ were calculated for each gastroenterologist, using data from the entire colon and then for each colonic segment separately. Actual ADR was compared with estimated ADR based on the measured APDRQ.RESULTS:During 1,921 colonoscopies, 2,285 polyps were removed; 1,122 (49%) were adenomas. The mean (s.d.) PDR for the group was 49% (12.4%) (range, 16–64%). The mean (s.d.) ADR was 31% (7.4%) (range, 13–42%). PDR and ADR correlated well in segments proximal to the splenic flexure, but diverged in distal segments. ADR was significantly higher in the right colon (17.1%) than in the left (13.5%) (P=0.001). The correlation between estimated and actual ADR using the APDRQ was significantly higher in the right colon (r=0.95 (95% confidence interval (CI), 0.87–0.98)) than in the left (r=0.59 (95% CI, 0.17–0.83)) (P<0.05).CONCLUSIONS:Although PDR and ADR correlate well in segments proximal to the splenic flexure, they do not correlate well in the left colon. Caution should be exercised when using PDR as a surrogate for ADR if data from the rectum and sigmoid are included.

Performance of the patency capsule compared with nonenteroclysis radiologic examinations in patients with known or suspected intestinal strictures

BACKGROUND The patency capsule (PC) is used before capsule endoscopy (CE) in patients with known or suspected small-bowel (SB) strictures or obstruction (SBO) to avoid CE retention. False-positive PC examination results can occur in patients with delayed transit without obstruction, precluding the use of CE. Radiological tests are another option to evaluate the presence of SBO before CE. OBJECTIVES Comparison of the PC and radiological examinations to detect clinically significant SB strictures. MAIN OUTCOME MEASUREMENTS Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the PC, and radiological tests for detecting significant strictures. RESULTS Forty-two patients underwent a PC study and radiological examinations. Both of the examinations showed similar sensitivity (57% vs 71%; P = 1.00) and specificity (86% vs 97%; P = .22). The receiver-operating characteristic curves evaluating combined sensitivity and specificity were also similar in both the PC and radiological examinations (0.71 vs 0.84, respectively; P = .46). Pooling results from both the PC and radiological tests had the highest sensitivity and NPV (100%, 100%). False-positive results occurred in 5 PC examinations and 1 radiological examination. The PC examination had 3 false-negative results (9%), whereas radiological tests had 2 (6%). LIMITATIONS Retrospective study. CONCLUSIONS The NPV for the PC and radiological tests were not significantly different, suggesting that if findings on either test are negative before CE, the patient will most likely pass the capsule without incident. Radiological tests can be used to minimize PC study false-positive results by confirming or excluding the presence of a significant stricture suspected by the PC and to localize the PC if passage is delayed.

Automated Polyp Detection in Colonoscopy Videos Using Shape and Context Information

This paper presents the culmination of our research in designing a system for computer-aided detection (CAD) of polyps in colonoscopy videos. Our system is based on a hybrid context-shape approach, which utilizes context information to remove non-polyp structures and shape information to reliably localize polyps. Specifically, given a colonoscopy image, we first obtain a crude edge map. Second, we remove non-polyp edges from the edge map using our unique feature extraction and edge classification scheme. Third, we localize polyp candidates with probabilistic confidence scores in the refined edge maps using our novel voting scheme. The suggested CAD system has been tested using two public polyp databases, CVC-ColonDB, containing 300 colonoscopy images with a total of 300 polyp instances from 15 unique polyps, and ASU-Mayo database, which is our collection of colonoscopy videos containing 19,400 frames and a total of 5,200 polyp instances from 10 unique polyps. We have evaluated our system using free-response receiver operating characteristic (FROC) analysis. At 0.1 false positives per frame, our system achieves a sensitivity of 88.0% for CVC-ColonDB and a sensitivity of 48% for the ASU-Mayo database. In addition, we have evaluated our system using a new detection latency analysis where latency is defined as the time from the first appearance of a polyp in the colonoscopy video to the time of its first detection by our system. At 0.05 false positives per frame, our system yields a polyp detection latency of 0.3 seconds.