Benchmarking definitions of false-positive alerts during computer-aided polyp detection in colonoscopy

Assessment of the role of false-positive alerts in computer-aided polyp detection for assistance capabilities

BACKGROUND AND AIM

False positives (FPs) pose a significant challenge in the application of artificial intelligence (AI) for polyp detection during colonoscopy. The study aimed to quantitatively evaluate the impact of computer-aided polyp detection (CADe) systems' FPs on endoscopists.

METHODS

The model's FPs were categorized into four gradients: 0-5, 5-10, 10-15, and 15-20 FPs per minute (FPPM). Fifty-six colonoscopy videos were collected for a crossover study involving 10 endoscopists. Polyp missed rate (PMR) was set as primary outcome. Subsequently, to further verify the impact of FPPM on the assistance capability of AI in clinical environments, a secondary analysis was conducted on a prospective randomized controlled trial (RCT) from Renmin Hospital of Wuhan University in China from July 1 to October 15, 2020, with the adenoma detection rate (ADR) as primary outcome.

RESULTS

Compared with routine group, CADe reduced PMR when FPPM was less than 5. However, with the continuous increase of FPPM, the beneficial effect of CADe gradually weakens. For secondary analysis of RCT, a total of 956 patients were enrolled. In AI-assisted group, ADR is higher when FPPM ≤ 5 compared with FPPM > 5 (CADe group: 27.78% vs 11.90%; P = 0.014; odds ratio [OR], 0.351; 95% confidence interval [CI], 0.152-0.812; COMBO group: 38.40% vs 23.46%, P = 0.029; OR, 0.427; 95% CI, 0.199-0.916). After AI intervention, ADR increased when FPPM ≤ 5 (27.78% vs 14.76%; P = 0.001; OR, 0.399; 95% CI, 0.231-0.690), but no statistically significant difference was found when FPPM > 5 (11.90% vs 14.76%, P = 0.788; OR, 1.111; 95% CI, 0.514-2.403).

CONCLUSION

The level of FPs of CADe does affect its effectiveness as an aid to endoscopists, with its best effect when FPPM is less than 5.

Evaluating false-positive detection in a computer-aided detection system for colonoscopy

BACKGROUND AND AIM

Computer-aided detection (CADe) systems can efficiently detect polyps during colonoscopy. However, false-positive (FP) activation is a major limitation of CADe. We aimed to compare the rate and causes of FP using CADe before and after an update designed to reduce FP.

METHODS

We analyzed CADe-assisted colonoscopy videos recorded between July 2022 and October 2022. The number and causes of FPs and excessive time spent by the endoscopist on FP (ET) were compared pre- and post-update using 1:1 propensity score matching.

RESULTS

During the study period, 191 colonoscopy videos (94 and 97 in the pre- and post-update groups, respectively) were recorded. Propensity score matching resulted in 146 videos (73 in each group). The mean number of FPs and median ET per colonoscopy were significantly lower in the post-update group than those in the pre-update group (4.2 ± 3.7 vs 18.1 ± 11.1; P < 0.001 and 0 vs 16 s; P < 0.001, respectively). Mucosal tags, bubbles, and folds had the strongest association with decreased FP post-update (pre-update vs post-update: 4.3 ± 3.6 vs 0.4 ± 0.8, 0.32 ± 0.70 vs 0.04 ± 0.20, and 8.6 ± 6.7 vs 1.6 ± 1.7, respectively). There was no significant decrease in the true positive rate (post-update vs pre-update: 95.0% vs 99.2%; P = 0.09) or the adenoma detection rate (post-update vs pre-update: 52.1% vs 49.3%; P = 0.87).

CONCLUSIONS

The updated CADe can reduce FP without impairing polyp detection. A reduction in FP may help relieve the burden on endoscopists.

Single Versus Second Observer vs Artificial Intelligence to Increase the ADENOMA Detection Rate of Colonoscopy-A Network Analysis

BACKGROUND AND AIMS

Screening colonoscopy has significantly contributed to the reduction of the incidence of colorectal cancer (CRC) and its associated mortality, with adenoma detection rate (ADR) as the quality marker. To increase the ADR, various solutions have been proposed including the utilization of Artificial Intelligence (AI) and employing second observers during colonoscopies. In the interest of AI improving ADR independently, without a second observer, and the operational similarity between AI and second observer, this network meta-analysis aims at evaluating the effectiveness of AI, second observer, and a single observer in improving ADR.

METHODS

We searched the Medline, Embase, Cochrane, Web of Science Core Collection, Korean Citation Index, SciELO, Global Index Medicus, and Cochrane. A direct head-to-head comparator analysis and network meta-analysis were performed using the random-effects model. The odds ratio (OR) was calculated with a 95% confidence interval (CI) and p-value < 0.05 was considered statistically significant.

RESULTS

We analyzed 26 studies, involving 22,560 subjects. In the direct comparative analysis, AI demonstrated higher ADR (OR: 0.668, 95% CI 0.595-0.749, p < 0.001) than single observer. Dual observer demonstrated a higher ADR (OR: 0.771, 95% CI 0.688-0.865, p < 0.001) than single operator. In network meta-analysis, results were consistent on the network meta-analysis, maintaining consistency. No statistical difference was noted when comparing AI to second observer. (RR 1.1 (0.9-1.2, p = 0.3). Results were consistent when evaluating only RCTs. Net ranking provided higher score to AI followed by second observer followed by single observer.

CONCLUSION

Artificial Intelligence and second-observer colonoscopy showed superior success in Adenoma Detection Rate when compared to single-observer colonoscopy. Although not statistically significant, net ranking model favors the superiority of AI to the second observer.

Improvement of adenoma detection rate by two computer-aided colonic polyp detection systems in high adenoma detectors: a randomized multicenter trial

BACKGROUND

This study aimed to evaluate the benefits of a self-developed computer-aided polyp detection system (SD-CADe) and a commercial system (CM-CADe) for high adenoma detectors compared with white-light endoscopy (WLE) as a control.

METHODS

Average-risk 50-75-year-old individuals who underwent screening colonoscopy at five referral centers were randomized to SD-CADe, CM-CADe, or WLE groups (1:1:1 ratio). Trainees and staff with an adenoma detection rate (ADR) of ≥35% were recruited. The primary outcome was ADR. Secondary outcomes were the proximal adenoma detection rate (pADR), advanced adenoma detection rate (AADR), and the number of adenomas, proximal adenomas, and advanced adenomas per colonoscopy (APC, pAPC, and AAPC, respectively).

RESULTS

The study enrolled 1200 participants. The ADR in the control, CM-CADe, and SD-CADe groups was 38.3%, 50.0%, and 54.8%, respectively. The pADR was 23.0%, 32.3%, and 38.8%, respectively. AADR was 6.0%, 10.3%, and 9.5%, respectively. After adjustment, the ADR and pADR in both intervention groups were significantly higher than in controls (all P<0.05). The APC in the control, CM-CADe, and SD-CADe groups was 0.66, 1.04, and 1.16, respectively. The pAPC was 0.33, 0.53, and 0.64, respectively, and the AAPC was 0.07, 0.12, and 0.10, respectively. Both CADe systems showed significantly higher APC and pAPC than WLE. AADR and AAPC were improved in both CADe groups versus control, although the differences were not statistically significant.

CONCLUSION

Even in high adenoma detectors, CADe significantly improved ADR and APC. The AADR tended to be higher with both systems, and this may enhance colorectal cancer prevention.

Performance comparison between two computer-aided detection colonoscopy models by trainees using different false positive thresholds: a cross-sectional study in Thailand

BACKGROUND/AIMS

This study aims to compare polyp detection performance of "Deep-GI," a newly developed artificial intelligence (AI) model, to a previously validated AI model computer-aided polyp detection (CADe) using various false positive (FP) thresholds and determining the best threshold for each model.

METHODS

Colonoscopy videos were collected prospectively and reviewed by three expert endoscopists (gold standard), trainees, CADe (CAD EYE; Fujifilm Corp.), and Deep-GI. Polyp detection sensitivity (PDS), polyp miss rates (PMR), and false-positive alarm rates (FPR) were compared among the three groups using different FP thresholds for the duration of bounding boxes appearing on the screen.

RESULTS

In total, 170 colonoscopy videos were used in this study. Deep-GI showed the highest PDS (99.4% vs. 85.4% vs. 66.7%, p<0.01) and the lowest PMR (0.6% vs. 14.6% vs. 33.3%, p<0.01) when compared to CADe and trainees, respectively. Compared to CADe, Deep-GI demonstrated lower FPR at FP thresholds of ≥0.5 (12.1 vs. 22.4) and ≥1 second (4.4 vs. 6.8) (both p<0.05). However, when the threshold was raised to ≥1.5 seconds, the FPR became comparable (2 vs. 2.4, p=0.3), while the PMR increased from 2% to 10%.

CONCLUSION

Compared to CADe, Deep-GI demonstrated a higher PDS with significantly lower FPR at ≥0.5- and ≥1-second thresholds. At the ≥1.5-second threshold, both systems showed comparable FPR with increased PMR.

Performance evaluation of a computer-aided polyp detection system with artificial intelligence for colonoscopy

OBJECTIVES

A computer-aided detection (CAD) system was developed to support the detection of colorectal lesions by deep learning using video images of lesions and normal mucosa recorded during colonoscopy. The study's purpose was to evaluate the stand-alone performance of this device under blinded conditions.

METHODS

This multicenter prospective observational study was conducted at four Japanese institutions. We used 326 videos of colonoscopies recorded with patient consent at institutions in which the Ethics Committees approved the study. The sensitivity of successful detection of the CAD system was calculated using the target lesions, which were detected by adjudicators from two facilities for each lesion appearance frame; inconsistencies were settled by consensus. Successful detection was defined as display of the detection flag on the lesion for more than 0.5 s within 3 s of appearance.

RESULTS

Of the 556 target lesions from 185 cases, detection success sensitivity was 97.5% (95% confidence interval [CI] 95.8-98.5%). The "successful detection sensitivity per colonoscopy" was 93% (95% CI 88.3-95.8%). For the frame-based sensitivity, specificity, positive predictive value, and negative predictive value were 86.6% (95% CI 84.8-88.4%), 84.7% (95% CI 83.8-85.6%), 34.9% (95% CI 32.3-37.4%), and 98.2% (95% CI 97.8-98.5%), respectively.

TRIAL REGISTRATION

University Hospital Medical Information Network (UMIN000044622).

The Role of Artificial Intelligence in Colorectal Cancer Screening: Lesion Detection and Lesion Characterization

Colorectal cancer remains a leading cause of cancer-related morbidity and mortality worldwide, despite the widespread uptake of population surveillance strategies. This is in part due to the persistent development of 'interval colorectal cancers', where patients develop colorectal cancer despite appropriate surveillance intervals, implying pre-malignant polyps were not resected at a prior colonoscopy. Multiple techniques have been developed to improve the sensitivity and accuracy of lesion detection and characterisation in an effort to improve the efficacy of colorectal cancer screening, thereby reducing the incidence of interval colorectal cancers. This article presents a comprehensive review of the transformative role of artificial intelligence (AI), which has recently emerged as one such solution for improving the quality of screening and surveillance colonoscopy. Firstly, AI-driven algorithms demonstrate remarkable potential in addressing the challenge of overlooked polyps, particularly polyp subtypes infamous for escaping human detection because of their inconspicuous appearance. Secondly, AI empowers gastroenterologists without exhaustive training in advanced mucosal imaging to characterise polyps with accuracy similar to that of expert interventionalists, reducing the dependence on pathologic evaluation and guiding appropriate resection techniques or referrals for more complex resections. AI in colonoscopy holds the potential to advance the detection and characterisation of polyps, addressing current limitations and improving patient outcomes. The integration of AI technologies into routine colonoscopy represents a promising step towards more effective colorectal cancer screening and prevention.

A Review of the Technology, Training, and Assessment Methods for the First Real-Time AI-Enhanced Medical Device for Endoscopy

Artificial intelligence (AI) has the potential to assist in endoscopy and improve decision making, particularly in situations where humans may make inconsistent judgments. The performance assessment of the medical devices operating in this context is a complex combination of bench tests, randomized controlled trials, and studies on the interaction between physicians and AI. We review the scientific evidence published about GI Genius, the first AI-powered medical device for colonoscopy to enter the market, and the device that is most widely tested by the scientific community. We provide an overview of its technical architecture, AI training and testing strategies, and regulatory path. In addition, we discuss the strengths and limitations of the current platform and its potential impact on clinical practice. The details of the algorithm architecture and the data that were used to train the AI device have been disclosed to the scientific community in the pursuit of a transparent AI. Overall, the first AI-enabled medical device for real-time video analysis represents a significant advancement in the use of AI for endoscopies and has the potential to improve the accuracy and efficiency of colonoscopy procedures.

Framework and metrics for the clinical use and implementation of artificial intelligence algorithms into endoscopy practice: recommendations from the American Society for Gastrointestinal Endoscopy Artificial Intelligence Task Force

In the past few years, we have seen a surge in the development of relevant artificial intelligence (AI) algorithms addressing a variety of needs in GI endoscopy. To accept AI algorithms into clinical practice, their effectiveness, clinical value, and reliability need to be rigorously assessed. In this article, we provide a guiding framework for all stakeholders in the endoscopy AI ecosystem regarding the standards, metrics, and evaluation methods for emerging and existing AI applications to aid in their clinical adoption and implementation. We also provide guidance and best practices for evaluation of AI technologies as they mature in the endoscopy space. Note, this is a living document; periodic updates will be published as progress is made and applications evolve in the field of AI in endoscopy.

Artificial intelligence empowers the second-observer strategy for colonoscopy: a randomized clinical trial

Background

In colonoscopy screening for colorectal cancer, human vision limitations may lead to higher miss rate of lesions; artificial intelligence (AI) assistance has been demonstrated to improve polyp detection. However, there still lacks direct evidence to demonstrate whether AI is superior to trainees or experienced nurses as a second observer to increase adenoma detection during colonoscopy. In this study, we aimed to compare the effectiveness of assistance from AI and human observer during colonoscopy.

Methods

A prospective multicenter randomized study was conducted from 2 September 2019 to 29 May 2020 at four endoscopy centers in China. Eligible patients were randomized to either computer-aided detection (CADe)-assisted group or observer-assisted group. The primary outcome was adenoma per colonoscopy (APC). Secondary outcomes included polyp per colonoscopy (PPC), adenoma detection rate (ADR), and polyp detection rate (PDR). We compared continuous variables and categorical variables by using R studio (version 3.4.4).

Results

A total of 1,261 (636 in the CADe-assisted group and 625 in the observer-assisted group) eligible patients were analysed. APC (0.42 vs 0.35, P = 0.034), PPC (1.13 vs 0.81, P < 0.001), PDR (47.5% vs 37.4%, P < 0.001), ADR (25.8% vs 24.0%, P = 0.464), the number of detected sessile polyps (683 vs 464, P < 0.001), and sessile adenomas (244 vs 182, P = 0.005) were significantly higher in the CADe-assisted group than in the observer-assisted group. False detections of the CADe system were lower than those of the human observer (122 vs 191, P < 0.001).

Conclusions

Compared with the human observer, the CADe system may improve the clinical outcome of colonoscopy and reduce disturbance to routine practice (Chictr.org.cn No.: ChiCTR1900025235).

Development and validation of a deep learning-based algorithm for colonoscopy quality assessment

BACKGROUND

Quality indicators should be assessed and monitored to improve colonoscopy quality in clinical practice. Endoscopists must enter relevant information in the endoscopy reporting system to facilitate data collection, which may be inaccurate. The current study aimed to develop a full deep learning-based algorithm to identify and analyze intra-procedural colonoscopy quality indicators based on endoscopy images obtained during the procedure.

METHODS

A deep learning system for classifying colonoscopy images for quality assurance purposes was developed and its performance was assessed with an independent dataset. The system was utilized to analyze captured images and results were compared with those of real-world reports.

RESULTS

In total, 10,417 images from the hospital endoscopy database and 3157 from Hyper-Kvasir open dataset were utilized to develop the quality assurance algorithm. The overall accuracy of the algorithm was 96.72% and that of the independent test dataset was 94.71%. Moreover, 761 real-world reports and colonoscopy images were analyzed. The accuracy of electronic reports about cecal intubation rate was 99.34% and that of the algorithm was 98.95%. The agreement rate for the assessment of polypectomy rates using the electronic reports and the algorithm was 0.87 (95% confidence interval 0.83-0.90). A good correlation was found between the withdrawal time calculated using the algorithm and that entered by the physician (correlation coefficient r = 0.959, p < 0.0001).

CONCLUSION

We proposed a novel deep learning-based algorithm that used colonoscopy images for quality assurance purposes. This model can be used to automatically assess intra-procedural colonoscopy quality indicators in clinical practice.

Deep Learning Computer-aided Polyp Detection Reduces Adenoma Miss Rate: A United States Multi-center Randomized Tandem Colonoscopy Study (CADeT-CS Trial)

BACKGROUND & AIMS

Artificial intelligence-based computer-aided polyp detection (CADe) systems are intended to address the issue of missed polyps during colonoscopy. The effect of CADe during screening and surveillance colonoscopy has not previously been studied in a United States (U.S.) population.

METHODS

We conducted a prospective, multi-center, single-blind randomized tandem colonoscopy study to evaluate a deep-learning based CADe system (EndoScreener, Shanghai Wision AI, China). Patients were enrolled across 4 U.S. academic medical centers from 2019 through 2020. Patients presenting for colorectal cancer screening or surveillance were randomized to CADe colonoscopy first or high-definition white light (HDWL) colonoscopy first, followed immediately by the other procedure in tandem fashion by the same endoscopist. The primary outcome was adenoma miss rate (AMR), and secondary outcomes included sessile serrated lesion (SSL) miss rate and adenomas per colonoscopy (APC).

RESULTS

A total of 232 patients entered the study, with 116 patients randomized to undergo CADe colonoscopy first and 116 patients randomized to undergo HDWL colonoscopy first. After the exclusion of 9 patients, the study cohort included 223 patients. AMR was lower in the CADe-first group compared with the HDWL-first group (20.12% [34/169] vs 31.25% [45/144]; odds ratio [OR], 1.8048; 95% confidence interval [CI], 1.0780-3.0217; P = .0247). SSL miss rate was lower in the CADe-first group (7.14% [1/14]) vs the HDWL-first group (42.11% [8/19]; P = .0482). First-pass APC was higher in the CADe-first group (1.19 [standard deviation (SD), 2.03] vs 0.90 [SD, 1.55]; P = .0323). First-pass ADR was 50.44% in the CADe-first group and 43.64 % in the HDWL-first group (P = .3091).

CONCLUSION

In this U.S. multicenter tandem colonoscopy randomized controlled trial, we demonstrate a decrease in AMR and SSL miss rate and an increase in first-pass APC with the use of a CADe-system when compared with HDWL colonoscopy alone.

Development and evaluation of a deep learning model to improve the usability of polyp detection systems during interventions

Background

The efficiency of artificial intelligence as computer‐aided detection (CADe) systems for colorectal polyps has been demonstrated in several randomized trials. However, CADe systems generate many distracting detections, especially during interventions such as polypectomies. Those distracting CADe detections are often induced by the introduction of snares or biopsy forceps as the systems have not been trained for such situations. In addition, there are a significant number of non‐false but not relevant detections, since the polyp has already been previously detected. All these detections have the potential to disturb the examiner's work.

Objectives

Development and evaluation of a convolutional neuronal network that recognizes instruments in the endoscopic image, suppresses distracting CADe detections, and reliably detects endoscopic interventions.

Methods

A total of 580 different examination videos from 9 different centers using 4 different processor types were screened for instruments and represented the training dataset (519,856 images in total, 144,217 contained a visible instrument). The test dataset included 10 full‐colonoscopy videos that were analyzed for the recognition of visible instruments and detections by a commercially available CADe system (GI Genius, Medtronic).

Results

The test dataset contained 153,623 images, 8.84% of those presented visible instruments (12 interventions, 19 instruments used). The convolutional neuronal network reached an overall accuracy in the detection of visible instruments of 98.59%. Sensitivity and specificity were 98.55% and 98.92%, respectively. A mean of 462.8 frames containing distracting CADe detections per colonoscopy were avoided using the convolutional neuronal network. This accounted for 95.6% of all distracting CADe detections.

Conclusions

Detection of endoscopic instruments in colonoscopy using artificial intelligence technology is reliable and achieves high sensitivity and specificity. Accordingly, the new convolutional neuronal network could be used to reduce distracting CADe detections during endoscopic procedures. Thus, our study demonstrates the great potential of artificial intelligence technology beyond mucosal assessment.

Comparing the number and relevance of false activations between 2 artificial intelligence computer-aided detection systems: the NOISE study

BACKGROUND AND AIMS

Artificial intelligence has been shown to be effective in polyp detection, and multiple computer-aided detection (CADe) systems have been developed. False-positive (FP) activation emerged as a possible way to benchmark CADe performance in clinical practice. The aim of this study was to validate a previously developed classification of FPs comparing the performances of different brands of approved CADe systems.

METHODS

We compared 2 different consecutive video libraries (40 video per arm) collected at Humanitas Research Hospital with 2 different CADe system brands (CADe A and CADe B). For each video, the number of CADe false activations, cause, and time spent by the endoscopist to examine the area erroneously highlighted were reported. The FP activations were classified according to the previously developed classification of FPs (the NOISE classification) according to their cause and relevance.

RESULTS

In CADe A 1021 FP activations were registered across the 40 videos (25.5 ± 12.2 FPs per colonoscopy), whereas in CADe B 1028 were identified (25.7 ± 13.2 FPs per colonoscopy; P = .53). Among them, 22.9 ± 9.9 (89.8% in CADe A) and 22.1 ± 10.0 (86.0% in CADe B) were because of artifacts from the bowel wall. Conversely, 2.6 ± 1.9 (10.2% in CADe A) and 3.5 ± 2.1 (14% in CADe B) were caused by bowel content (P = .45). Within CADe A each false activation required .2 ± .9 seconds, with 1.6 ± 1.0 FPs (6.3%) requiring additional time for endoscopic assessment. Comparable results were reported within CADe B with .2 ± .8 seconds spent per false activation and 1.8 ± 1.2 FPs per colonoscopy requiring additional inspection.

CONCLUSIONS

The use of a standardized nomenclature provided comparable results with either of the 2 recently approved CADe systems. (Clinical trial registration number: NCT04399590.).

Discovering the first US FDA-approved computer-aided polyp detection system

Colorectal cancer is the third most common cancer worldwide. Because of the slow progression of the precancerous precursors, an efficient endoscopic surveillance strategy may be expected. It seems that around one-fourth of colorectal malignancies are still missed during colonoscopy. Several endoscopic technologies have been introduced, without radical changes. Interest in the development of artificial intelligence applications in the medical field has grown in the past decade. Artificial intelligence can help to highlight a specific region of interest that needs closer examination for the identification of polyps. The aim of this review is to report the first clinical experiences with the first US FDA-approved, real-time, deep-learning, computer-aided detection system (GI Genius™, Medtronic).

Artificial intelligence-assisted colonoscopy: a narrative review of current data and clinical applications

Colonoscopy is the reference standard procedure for the prevention and diagnosis of colorectal cancer, which is a leading cause of cancer-related deaths in Singapore. Artificial intelligence systems are automated, objective and reproducible. Artificial intelligence-assisted colonoscopy has recently been introduced into clinical practice as a clinical decision support tool. This review article provides a summary of the current published data and discusses ongoing research and current clinical applications of artificial intelligence-assisted colonoscopy.

Computer-Aided Detection False Positives in Colonoscopy

Randomized control trials and meta-analyses comparing colonoscopies with and without computer-aided detection (CADe) assistance showed significant increases in adenoma detection rates (ADRs) with CADe. A major limitation of CADe is its false positives (FPs), ranked 3rd in importance among 59 research questions in a modified Delphi consensus review. The definition of FPs varies. One commonly used definition defines an FP as an activation of the CADe system, irrespective of the number of frames or duration of time, not due to any polypoid or nonpolypoid lesions. Although only 0.07 to 0.2 FPs were observed per colonoscopy, video analysis studies using FPs as the primary outcome showed much higher numbers of 26 to 27 per colonoscopy. Most FPs were of short duration (91% < 0.5 s). A higher number of FPs was also associated with suboptimal bowel preparation. The appearance of FPs can lead to user fatigue. The polypectomy of FPs results in increased procedure time and added use of resources. Re-training the CADe algorithms is one way to reduce FPs but is not practical in the clinical setting during colonoscopy. Water exchange (WE) is an emerging method that the colonoscopist can use to provide salvage cleaning during insertion. We discuss the potential of WE for reducing FPs as well as the augmentation of ADRs through CADe.