Artificial intelligence in colonoscopy: A review on the current status

The imitation game: a review of the use of artificial intelligence in colonoscopy, and endoscopists' perceptions thereof

The development of deep learning systems in artificial intelligence (AI) has enabled advances in endoscopy, and AI-aided colonoscopy has recently been ushered into clinical practice as a clinical decision-support tool. This has enabled real-time AI-aided detection of polyps with a higher sensitivity than the average endoscopist, and evidence to support its use has been promising thus far. This review article provides a summary of currently published data relating to AI-aided colonoscopy, discusses current clinical applications, and introduces ongoing research directions. We also explore endoscopists' perceptions and attitudes toward the use of this technology, and discuss factors influencing its uptake in clinical practice.

Effect of real-time computer-aided detection of colorectal adenoma in routine colonoscopy (COLO-GENIUS): a single-centre randomised controlled trial

BACKGROUND

Artificial intelligence systems have been developed to improve polyp detection. We aimed to evaluate the effect of real-time computer-aided detection (CADe) on the adenoma detection rate (ADR) in routine colonoscopy.

METHODS

This single-centre randomised controlled trial (COLO-GENIUS) was done at the Digestive Endoscopy Unit, Pôle Digestif Paris-Bercy, Clinique Paris-Bercy, Charenton-le-Pont, France. All consecutive individuals aged 18 years or older who were scheduled for a total colonoscopy and had an American Society of Anesthesiologists score of 1-3 were screened for inclusion. After the caecum was reached and the colonic preparation was appropriate, eligible participants were randomly assigned (1:1; computer-generated random numbers list) to either standard colonoscopy or CADe-assisted colonoscopy (GI Genius 2.0.2; Medtronic). Participants and cytopathologists were masked to study assignment, whereas endoscopists were not. The primary outcome was ADR, which was assessed in the modified intention-to-treat population (all randomly assigned participants except those with misplaced consent forms). Safety was analysed in all included patients. According to statistical calculations, 20 endoscopists from the Clinique Paris-Bercy had to include approximately 2100 participants with 1:1 randomisation. The trial is complete and registered with ClinicalTrials.gov, NCT04440865.

FINDINGS

Between May 1, 2021, and May 1, 2022, 2592 participants were assessed for eligibility, of whom 2039 were randomly assigned to standard colonoscopy (n=1026) or CADe-assisted colonoscopy (n=1013). 14 participants in the standard group and ten participants in the CADe group were then excluded due to misplaced consent forms, leaving 2015 participants (979 [48·6%] men and 1036 [51·4%] women) in the modified intention-to-treat analysis. ADR was 33·7% (341 of 1012 colonoscopies) in the standard group and 37·5% (376 of 1003 colonoscopies) in the CADe group (estimated mean absolute difference 4·1 percentage points [95% CI 0·0-8·1]; p=0·051). One bleeding event without deglobulisation occurred in the CADe group after a large (>2 cm) polyp resection and resolved after a haemostasis clip was placed during a second colonoscopy.

INTERPRETATION

Our findings support the benefits of CADe, even in a non-academic centre. Systematic use of CADe in routine colonoscopy should be considered.

FUNDING

None.

EnsemDeepCADx: Empowering Colorectal Cancer Diagnosis with Mixed-Dataset Features and Ensemble Fusion CNNs on Evidence-Based CKHK-22 Dataset

Colorectal cancer is associated with a high mortality rate and significant patient risk. Images obtained during a colonoscopy are used to make a diagnosis, highlighting the importance of timely diagnosis and treatment. Using techniques of deep learning could enhance the diagnostic accuracy of existing systems. Using the most advanced deep learning techniques, a brand-new EnsemDeepCADx system for accurate colorectal cancer diagnosis has been developed. The optimal accuracy is achieved by combining Convolutional Neural Networks (CNNs) with transfer learning via bidirectional long short-term memory (BILSTM) and support vector machines (SVM). Four pre-trained CNN models comprise the ADaDR-22, ADaR-22, and DaRD-22 ensemble CNNs: AlexNet, DarkNet-19, DenseNet-201, and ResNet-50. In each of its stages, the CADx system is thoroughly evaluated. From the CKHK-22 mixed dataset, colour, greyscale, and local binary pattern (LBP) image datasets and features are utilised. In the second stage, the returned features are compared to a new feature fusion dataset using three distinct CNN ensembles. Next, they incorporate ensemble CNNs with SVM-based transfer learning by comparing raw features to feature fusion datasets. In the final stage of transfer learning, BILSTM and SVM are combined with a CNN ensemble. The testing accuracy for the ensemble fusion CNN DarD-22 using BILSTM and SVM on the original, grey, LBP, and feature fusion datasets was optimal (95.96%, 88.79%, 73.54%, and 97.89%). Comparing the outputs of all four feature datasets with those of the three ensemble CNNs at each stage enables the EnsemDeepCADx system to attain its highest level of accuracy.

The Impacts of Computer-Aided Detection of Colorectal Polyps on Subsequent Colonoscopy Surveillance Intervals: Simulation Study

BACKGROUND

Computer-aided detection (CADe) of colorectal polyps has been shown to increase adenoma detection rates, which would potentially shorten subsequent surveillance intervals.

OBJECTIVE

The purpose of this study is to simulate the potential changes in subsequent colonoscopy surveillance intervals after the application of CADe in a large cohort of patients.

METHODS

We simulated the projected increase in polyp and adenoma detection by universal CADe application in our patients who had undergone colonoscopy with complete endoscopic and histological findings between 2016 and 2020. The simulation was based on bootstrapping the published performance of CADe. The corresponding changes in surveillance intervals for each patient, as recommended by the US Multi-Society Task Force on Colorectal Cancer (USMSTF) or the European Society of Gastrointestinal Endoscopy (ESGE), were determined after the CADe was determined.

RESULTS

A total of 3735 patients who had undergone colonoscopy were included. Based on the simulated CADe effect, the application of CADe would result in 19.1% (n=714) and 1.9% (n=71) of patients having shorter surveillance intervals, according to the USMSTF and ESGE guidelines, respectively. In particular, all (or 2.7% (n=101) of the total) patients who were originally scheduled to have 3-5 years of surveillance would have their surveillance intervals shortened to 3 years, following the USMSTF guidelines. The changes in this group of patients were largely attributed to an increase in the number of adenomas (n=75, 74%) rather than serrated lesions being detected.

CONCLUSIONS

Widespread adoption of CADe would inevitably increase the demand for surveillance colonoscopies with the shortening of original surveillance intervals, particularly following the current USMSTF guideline.

Experimental evidence of effective human-AI collaboration in medical decision-making

Artificial Intelligence (AI) systems are precious support for decision-making, with many applications also in the medical domain. The interaction between MDs and AI enjoys a renewed interest following the increased possibilities of deep learning devices. However, we still have limited evidence-based knowledge of the context, design, and psychological mechanisms that craft an optimal human-AI collaboration. In this multicentric study, 21 endoscopists reviewed 504 videos of lesions prospectively acquired from real colonoscopies. They were asked to provide an optical diagnosis with and without the assistance of an AI support system. Endoscopists were influenced by AI ([Formula: see text]), but not erratically: they followed the AI advice more when it was correct ([Formula: see text]) than incorrect ([Formula: see text]). Endoscopists achieved this outcome through a weighted integration of their and the AI opinions, considering the case-by-case estimations of the two reliabilities. This Bayesian-like rational behavior allowed the human-AI hybrid team to outperform both agents taken alone. We discuss the features of the human-AI interaction that determined this favorable outcome.