A video based benchmark data set (ENDOTEST) to evaluate computer-aided polyp detection systems

GI genius endoscopy module: a clinical profile

INTRODUCTION

The identification of early-stage colorectal cancers (CRC) and the resection of pre-cancerous neoplastic lesions through colonoscopy allows to decrease both CRC incidence and mortality. However, colonoscopy miss rates up to 26% for adenomas and 9% for advanced adenomas have been reported. In recent years, artificial intelligence (AI) systems have been emerging as easy-to-use tools, potentially lowering the risk of missing lesions.

AREAS COVERED

This review paper focuses on GI Genius device (Medtronic Co. Minneapolis, MN, U.S.A.) a computer-assisted tool designed to assist endoscopists during standard white-light colonoscopies in detecting mucosal lesions.

EXPERT OPINION

Randomized controlled trials (RCTs) suggest that GI Genius is a safe and effective tool for improving adenoma detection, especially in CRC screening and surveillance colonoscopies. However, its impact seems to be less significant among experienced endoscopists and in real-world clinical scenarios compared to the controlled conditions of RCTs. Furthermore, it appears that GI Genius mainly enhances the detection of non-advanced, small polyps, but does not significantly impact the identification of advanced and difficult-to-detect adenoma. When using GI Genius, no complications were documented. Only a small number of studies reported an increased in withdrawal time or the removal of non-neoplastic lesions.

Direct comparison of multiple computer-aided polyp detection systems

BACKGROUND AND STUDY AIMS

 Artificial intelligence (AI)-based systems for computer-aided detection (CADe) of polyps receive regular updates and occasionally offer customizable detection thresholds, both of which impact their performance, but little is known about these effects. This study aimed to compare the performance of different CADe systems on the same benchmark dataset.

METHODS

 101 colonoscopy videos were used as benchmark. Each video frame with a visible polyp was manually annotated with bounding boxes, resulting in 129 705 polyp images. The videos were then analyzed by three different CADe systems, representing five conditions: two versions of GI Genius, Endo-AID with detection Types A and B, and EndoMind, a freely available system. Evaluation included an analysis of sensitivity and false-positive rate, among other metrics.

RESULTS

 Endo-AID detection Type A, the earlier version of GI Genius, and EndoMind detected all 93 polyps. Both the later version of GI Genius and Endo-AID Type B missed 1 polyp. The mean per-frame sensitivities were 50.63 % and 67.85 %, respectively, for the earlier and later versions of GI Genius, 65.60 % and 52.95 %, respectively, for Endo-AID Types A and B, and 60.22 % for EndoMind.

CONCLUSIONS

 This study compares the performance of different CADe systems, different updates, and different configuration modes. This might help clinicians to select the most appropriate system for their specific needs.

Sources of performance variability in deep learning-based polyp detection

PURPOSE

Validation metrics are a key prerequisite for the reliable tracking of scientific progress and for deciding on the potential clinical translation of methods. While recent initiatives aim to develop comprehensive theoretical frameworks for understanding metric-related pitfalls in image analysis problems, there is a lack of experimental evidence on the concrete effects of common and rare pitfalls on specific applications. We address this gap in the literature in the context of colon cancer screening.

METHODS

Our contribution is twofold. Firstly, we present the winning solution of the Endoscopy Computer Vision Challenge on colon cancer detection, conducted in conjunction with the IEEE International Symposium on Biomedical Imaging 2022. Secondly, we demonstrate the sensitivity of commonly used metrics to a range of hyperparameters as well as the consequences of poor metric choices.

RESULTS

Based on comprehensive validation studies performed with patient data from six clinical centers, we found all commonly applied object detection metrics to be subject to high inter-center variability. Furthermore, our results clearly demonstrate that the adaptation of standard hyperparameters used in the computer vision community does not generally lead to the clinically most plausible results. Finally, we present localization criteria that correspond well to clinical relevance.

CONCLUSION

We conclude from our study that (1) performance results in polyp detection are highly sensitive to various design choices, (2) common metric configurations do not reflect the clinical need and rely on suboptimal hyperparameters and (3) comparison of performance across datasets can be largely misleading. Our work could be a first step towards reconsidering common validation strategies in deep learning-based colonoscopy and beyond.

A Real-Time Polyp-Detection System with Clinical Application in Colonoscopy Using Deep Convolutional Neural Networks

Colorectal cancer (CRC) is a leading cause of cancer-related deaths worldwide. The best method to prevent CRC is with a colonoscopy. During this procedure, the gastroenterologist searches for polyps. However, there is a potential risk of polyps being missed by the gastroenterologist. Automated detection of polyps helps to assist the gastroenterologist during a colonoscopy. There are already publications examining the problem of polyp detection in the literature. Nevertheless, most of these systems are only used in the research context and are not implemented for clinical application. Therefore, we introduce the first fully open-source automated polyp-detection system scoring best on current benchmark data and implementing it ready for clinical application. To create the polyp-detection system (ENDOMIND-Advanced), we combined our own collected data from different hospitals and practices in Germany with open-source datasets to create a dataset with over 500,000 annotated images. ENDOMIND-Advanced leverages a post-processing technique based on video detection to work in real-time with a stream of images. It is integrated into a prototype ready for application in clinical interventions. We achieve better performance compared to the best system in the literature and score a F1-score of 90.24% on the open-source CVC-VideoClinicDB benchmark.

Detection of Colorectal Polyps from Colonoscopy Using Machine Learning: A Survey on Modern Techniques

Given the increased interest in utilizing artificial intelligence as an assistive tool in the medical sector, colorectal polyp detection and classification using deep learning techniques has been an active area of research in recent years. The motivation for researching this topic is that physicians miss polyps from time to time due to fatigue and lack of experience carrying out the procedure. Unidentified polyps can cause further complications and ultimately lead to colorectal cancer (CRC), one of the leading causes of cancer mortality. Although various techniques have been presented recently, several key issues, such as the lack of enough training data, white light reflection, and blur affect the performance of such methods. This paper presents a survey on recently proposed methods for detecting polyps from colonoscopy. The survey covers benchmark dataset analysis, evaluation metrics, common challenges, standard methods of building polyp detectors and a review of the latest work in the literature. We conclude this paper by providing a precise analysis of the gaps and trends discovered in the reviewed literature for future work.

Deep learning as a novel method for endoscopic diagnosis of chronic atrophic gastritis: a prospective nested case-control study

Background and aims

Chronic atrophic gastritis (CAG) is a precancerous disease that often leads to the development of gastric cancer (GC) and is positively correlated with GC morbidity. However, the sensitivity of the endoscopic diagnosis of CAG is only 42%. Therefore, we developed a real-time video monitoring model for endoscopic diagnosis of CAG based on U-Net deep learning (DL) and conducted a prospective nested case–control study to evaluate the diagnostic evaluation indices of the model and its consistency with pathological diagnosis.

Methods

Our cohort consisted of 1539 patients undergoing gastroscopy from December 1, 2020, to July 1, 2021. Based on pathological diagnosis, patients in the cohort were divided into the CAG group or the chronic nonatrophic gastritis (CNAG) group, and we assessed the diagnostic evaluation indices of this model and its consistency with pathological diagnosis after propensity score matching (PSM) to minimize selection bias in the study.

Results

After matching, the diagnostic evaluation indices and consistency evaluation of the model were better than those of endoscopists [sensitivity (84.02% vs. 62.72%), specificity (97.04% vs. 81.95%), positive predictive value (96.60% vs. 77.66%), negative predictive value (85.86% vs. 68.73%), accuracy rate (90.53% vs. 72.34%), Youden index (81.06% vs. 44.67%), odd product (172.5 vs. 7.64), positive likelihood ratio (28.39 vs. 3.47), negative likelihood ratio (0.16 vs. 0.45), AUC (95% CI) [0.909 (0.884–0.934) vs. 0.740 (0.702–0.778)] and Kappa (0.852 vs. 0.558)].

Conclusions

Our prospective nested case–control study proved that the diagnostic evaluation indices and consistency evaluation of the real-time video monitoring model for endoscopic diagnosis of CAG based on U-Net DL were superior to those of endoscopists.

Trial registrationChiCTR2100044458, 18/03/2020.