Difference between revisions of "Real-time automatic polyp detection system for colonoscopy using artificial intelligence"

From Clinfowiki
Jump to: navigation, search
Line 16: Line 16:
 
The first publication demonstrating accurate polyp detection in real time was by Urban et al.[5] They developed a CADe model retrospectively and reported an area under the curve of 0.991 (a measure in which values of 0.5 correspond to chance observation, and 1.0 is perfect accuracy) and an accuracy of 96%. A deep convolutional neural network (DCNN) is a type of deep learning model that is highly effective at performing image analysis. A polyp detection module also must process images at a minimum of 30 frames per second to be applicable during colonoscopy. They trained DCNNs to detect polyps using a diverse and representative set of 8,641 hand labeled images from screening colonoscopies collected from more than 2000 patients. The model was tested on 20 colonoscopy videos with a total duration of 5 hours. The CNN system detected and localized polyps well within real-time constraints using an ordinary desktop machine with a contemporary graphics processing unit. Their feasibility study suggested that CNN assistance during live colonoscopy will result in fewer missed polyps.[5]
 
The first publication demonstrating accurate polyp detection in real time was by Urban et al.[5] They developed a CADe model retrospectively and reported an area under the curve of 0.991 (a measure in which values of 0.5 correspond to chance observation, and 1.0 is perfect accuracy) and an accuracy of 96%. A deep convolutional neural network (DCNN) is a type of deep learning model that is highly effective at performing image analysis. A polyp detection module also must process images at a minimum of 30 frames per second to be applicable during colonoscopy. They trained DCNNs to detect polyps using a diverse and representative set of 8,641 hand labeled images from screening colonoscopies collected from more than 2000 patients. The model was tested on 20 colonoscopy videos with a total duration of 5 hours. The CNN system detected and localized polyps well within real-time constraints using an ordinary desktop machine with a contemporary graphics processing unit. Their feasibility study suggested that CNN assistance during live colonoscopy will result in fewer missed polyps.[5]
  
Wang et al[6] reported the first prospective randomized controlled trial examining an automatic polyp detection during colonoscopy and showed an increase of ADR by 50%, from 20% to 30%. This effect was mainly due to a higher rate of detection of small adenomas and hyperplastic polyps. Of the 1058 patients included, 536 were randomized to standard colonoscopy, and 522 were randomized to colonoscopy with computer-aided diagnosis. The artificial intelligence (AI) system significantly increased ADR (29.1%vs20.3%, p<0.001) and the mean number of adenomas per patient (0.53 vs 0.31, p<0.001). The real-time automatic polyp detection system (Shanghai Wision AI Co., Ltd.) was developed on a deep learning architecture with the aid of endoscopists and modelers. The detection delay was hardly noticeable for endoscopists.[6]
+
Wang et al [6] reported the first prospective randomized controlled trial examining an automatic polyp detection during colonoscopy and showed an increase of ADR by 50%, from 20% to 30%. This effect was mainly due to a higher rate of detection of small adenomas and hyperplastic polyps. Of the 1058 patients included, 536 were randomized to standard colonoscopy, and 522 were randomized to colonoscopy with computer-aided diagnosis. The artificial intelligence (AI) system significantly increased ADR (29.1%vs20.3%, p<0.001) and the mean number of adenomas per patient (0.53 vs 0.31, p<0.001). The real-time automatic polyp detection system (Shanghai Wision AI Co., Ltd.) was developed on a deep learning architecture with the aid of endoscopists and modelers. The detection delay was hardly noticeable for endoscopists.[6]
  
 
Klare et al [7] conducted a prospective evaluation of a CADe model based on hand-drafted features. Their model achieved a 29.1% ADR in 55 colonoscopies, by using the number of adenomas found by blinded, experienced endoscopists as a reference standard. They used KoloPol APDS, an automated polyp detection software (APDS) developed in Germany. It offers endoscopists visual clues by highlighting suspicious mucosal areas in a color-coded manner during colonoscopy.[7]
 
Klare et al [7] conducted a prospective evaluation of a CADe model based on hand-drafted features. Their model achieved a 29.1% ADR in 55 colonoscopies, by using the number of adenomas found by blinded, experienced endoscopists as a reference standard. They used KoloPol APDS, an automated polyp detection software (APDS) developed in Germany. It offers endoscopists visual clues by highlighting suspicious mucosal areas in a color-coded manner during colonoscopy.[7]
Line 27: Line 27:
  
 
'''References'''
 
'''References'''
1. https://www.medtronic.com/covidien/en-us/products/gastrointestinal-artificial-intelligence/gi-genius-intelligent-endoscopy.html
+
1.https://www.medtronic.com/covidien/en-us/products/gastrointestinal-artificial-intelligence/gi-genius-intelligent-endoscopy.html
 
2. Hassan C, Wallace MB, Sharma P, et al. New artificial intelligence system: first validation study versus experienced endoscopists for colorectal polyp detection. Gut. 2020;69(5):799-800. doi:10.1136/gutjnl-2019-319914
 
2. Hassan C, Wallace MB, Sharma P, et al. New artificial intelligence system: first validation study versus experienced endoscopists for colorectal polyp detection. Gut. 2020;69(5):799-800. doi:10.1136/gutjnl-2019-319914
 
3. Vinsard DG, Mori Y, Misawa M, et al. Quality assurance of computer-aided detection and diagnosis in colonoscopy. Gastrointest Endosc. 2019;90(1):55-63. doi:10.1016/j.gie.2019.03.019
 
3. Vinsard DG, Mori Y, Misawa M, et al. Quality assurance of computer-aided detection and diagnosis in colonoscopy. Gastrointest Endosc. 2019;90(1):55-63. doi:10.1016/j.gie.2019.03.019

Revision as of 03:14, 28 April 2020

Real-time automatic polyp detection system for colonoscopy using artificial intelligence

Introduction Real-time endoscopic image diagnosis support systems can now perform real-time detection of colon polyps during colonoscopy with the use of deep convolutional neural networks (DCCN) and hand-crafted AI algorithms. Medtronic, a global leader in medical technology launched the GI Genius™ Intelligent Endoscopy Module Artificial Intelligence System for Colonoscopy at the United European Gastroenterology Week in 2019.[1] Even though it is available in select European markets, it is not yet available in the US since FDA approval is pending. It is the first commercially available computer-aided detection (CADe) device in gastroenterology which can automatically highlight the presence of pre-cancerous lesions with a visual marker in real-time serving as an ever vigilant second observer.[1] The system works with all brands of colonoscopes and seamlessly integrates with the existing workflow. It works as an adjunct to the gastroenterologist during a colonoscopy and highlights regions with visual characteristics consistent with polyps of different morphologies, shapes and sizes.[1] AI algorithms for object detection usually comprise a DCCN trained using ground truth images annotated by experts. Once trained, these AI systems are able to detect and pinpoint objects in real time, such as colorectal polyps.[2] In contrast to radiological images which can be reviewed later, the endoscopic videos are not stored and have to be interpreted in real-time needing complex analysis of millions of frames even though the clinically relevant lesions are limited to only a few frames.[2]

How can CADe help in reducing incidence of colorectal cancer? Colorectal cancer (CRC) is the third most common form of cancer globally with 1.8 million new cases every year. Adenoma detection rate (ADR) for colonoscopists varies from 7% to 53% with substantial variability in miss rates for polyps. Unrecognized polyps within the visual field are an important problem to address. It is estimated that every 1% increase in ADR lowers the risk of interval colorectal cancers by 3%. Computer-assisted image analysis using DCCN (a deep learning model for image analysis) improves polyp detection, a surrogate of ADR. New strategies such as use of artificial intelligence are needed to increase the ADR and decrease performance variability during colonoscopy.

First Validation study of GI-Genius by Medtronic To improve colorectal polyp detection, the system was trained and validated on a dataset of white-light endoscopy videos. It included 1.5 million images from different perspectives taken from 840 patients. The 2684 histologically confirmed polyps were manually annotated by expert endoscopists.[2] These were from a high-quality randomized controlled trial and the study compared the detection rate and the reaction time (RT) on a lesion basis (n=337/338, sensitivity: 99.7%). False-positive frames were seen in less than 1% of frames from the whole colonoscopy. The RT was faster by the AI system as compared with endoscopists in 82% of cases (n=277/337; difference 1.27+3.81). The AI system was able to virtually detect all the lesions extracted from a high-quality randomized study performed by expert endoscopists with an anticipation of the diagnosis as compared with the human reader in the vast majority of the cases. They also showed that the rate of false-positive results is negligible, showing the high precision of an AI-based algorithm in discriminating between normal mucosa, on one hand, and adenomatous and serrated lesions, on the other. The nearly 100% sensitivity per lesions is clinically relevant, as it indicates that AI detection occurs irrespective of the shape, size, location and histology of the lesion.[2] The authors opined that the clinical implications of the findings are relevant when assuming that most of the adenoma miss rates at colonoscopy, as well as variability in adenoma detection rate across colonoscopists are related to perceptual errors. There is convincing evidence that individual endoscopists routinely fail to recognize polyps actually visible in the monitor. Limitations in human visual perception and other human biases, such as fatigue, distraction, lower levels of alertness during the examination increase such perceptual errors. AI appears to be the best way of mitigating them.[2]

CADe versus CADx There are two ways in which AI could be useful during colonoscopy. These include automated polyp detection (CADe) and automated polyp histology characterization (CADx).[3] This review focusses on CADe only. CADe minimizes the miss rates of polyps and increases the adenoma detection rate (ADR) and polyp detection rate (PDR).These help to decrease the interval colon cancer incidence. CADx is useful to predict the pathology of the polyp that is detected in real time to improve accuracy of optical biopsy. In vivo differentiation between neoplastic and non-neoplastic polyps by using endoscopic light properties without non-neoplastic polyp removal can reduce costs by avoiding histopathological examination by a pathologist.[4] This will also help to use the strategy of resect and discard. Medtronic GI-Genius currently focusses only on the CADe aspect.

Review of recent literature on CADe The first publication demonstrating accurate polyp detection in real time was by Urban et al.[5] They developed a CADe model retrospectively and reported an area under the curve of 0.991 (a measure in which values of 0.5 correspond to chance observation, and 1.0 is perfect accuracy) and an accuracy of 96%. A deep convolutional neural network (DCNN) is a type of deep learning model that is highly effective at performing image analysis. A polyp detection module also must process images at a minimum of 30 frames per second to be applicable during colonoscopy. They trained DCNNs to detect polyps using a diverse and representative set of 8,641 hand labeled images from screening colonoscopies collected from more than 2000 patients. The model was tested on 20 colonoscopy videos with a total duration of 5 hours. The CNN system detected and localized polyps well within real-time constraints using an ordinary desktop machine with a contemporary graphics processing unit. Their feasibility study suggested that CNN assistance during live colonoscopy will result in fewer missed polyps.[5]

Wang et al [6] reported the first prospective randomized controlled trial examining an automatic polyp detection during colonoscopy and showed an increase of ADR by 50%, from 20% to 30%. This effect was mainly due to a higher rate of detection of small adenomas and hyperplastic polyps. Of the 1058 patients included, 536 were randomized to standard colonoscopy, and 522 were randomized to colonoscopy with computer-aided diagnosis. The artificial intelligence (AI) system significantly increased ADR (29.1%vs20.3%, p<0.001) and the mean number of adenomas per patient (0.53 vs 0.31, p<0.001). The real-time automatic polyp detection system (Shanghai Wision AI Co., Ltd.) was developed on a deep learning architecture with the aid of endoscopists and modelers. The detection delay was hardly noticeable for endoscopists.[6]

Klare et al [7] conducted a prospective evaluation of a CADe model based on hand-drafted features. Their model achieved a 29.1% ADR in 55 colonoscopies, by using the number of adenomas found by blinded, experienced endoscopists as a reference standard. They used KoloPol APDS, an automated polyp detection software (APDS) developed in Germany. It offers endoscopists visual clues by highlighting suspicious mucosal areas in a color-coded manner during colonoscopy.[7]

Aziz et al [8] performed a systematic review and meta-analysis of the available studies to assess the impact of deep convoluted neural network (DCNN) based AI assisted colonoscopy in improving the ADR and PDR. A total of 3 studies with 2815 patients (1415 in standard colonoscopy (SC) group and 1400 in AI group) were included. AI colonoscopy resulted in significantly improved ADR (32.9% vs 20.8%, RR: 1.58, 95 % CI 1.39 – 1.80, p = <0.001) and PDR (43.0% vs 27.8%, RR: 1.55, 95 % CI 1.39 –1.72, p = <0.001) compared to SC.[8]

Future of real-time endoscopic image diagnosis support systems The field of real-time endoscopic image diagnosis support system is advancing rapidly and we need more randomized controlled studies to clarify the unresolved aspects. Other issues include standardization of outcomes, need for widespread dataset availability, more real-world applications of CADx to distinguish between various types of colon polyps, regulatory approvals, AI assessment of bowel preparation quality, lesion size measurement, morphology description, identification of lesion features associated with deep and superficial submucosal invasion of cancer, real-time guidance of therapeutic procedures and automated report generation.[3,4] Will the reliance on AI make the new generation of colonoscopists less skillful and meticulous given the security provided by these tools?[3] Only time will tell.


References 1.https://www.medtronic.com/covidien/en-us/products/gastrointestinal-artificial-intelligence/gi-genius-intelligent-endoscopy.html 2. Hassan C, Wallace MB, Sharma P, et al. New artificial intelligence system: first validation study versus experienced endoscopists for colorectal polyp detection. Gut. 2020;69(5):799-800. doi:10.1136/gutjnl-2019-319914 3. Vinsard DG, Mori Y, Misawa M, et al. Quality assurance of computer-aided detection and diagnosis in colonoscopy. Gastrointest Endosc. 2019;90(1):55-63. doi:10.1016/j.gie.2019.03.019 4. Kudo SE, Mori Y, Misawa M, et al. Artificial intelligence and colonoscopy: Current status and future perspectives. DIG. ENDOSC.. 2019;31(4):363-371. doi:10.1111/den.13340 5. Urban G, Tripathi P, Alkayali T, et al. Deep Learning Localizes and Identifies Polyps in Real Time With 96% Accuracy in Screening Colonoscopy. Gastroenterology. 2018 Oct;155(4):1069-1078.e8. doi: 10.1053/j.gastro.2018.06.037 6. Wang P, Berzin TM, Glissen Brown JR, et al. Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study. Gut. 2019;68(10):1813-1819. doi:10.1136/gutjnl-2018-317500 7. Klare P, Sander C, Prinzen M, et al. Automated polyp detection in the colorectum: a prospective study (with videos). Gastrointest Endosc. 2019;89(3):576-582.e1. doi:10.1016/j.gie.2018.09.042 8. Aziz M, Fatima R, Dong C, Lee-Smith W, Nawras A. The Impact of Deep Convoluted Neural Network based Artificial Intelligence on Colonoscopy Outcomes: A Systematic Review with Meta-analysis. J Gastroenterol Hepatol. 2020;. doi:10.1111/jgh.15070

Submitted by (Suku George MD MPH)