FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2968146089> ?p ?o ?g. }

• W2968146089 abstract "Introduction Artificial intelligence (AI) can potentially improve adenoma detection rates. Previous work focussed on still images and selected video sequences which may be subject to bias and lack clinical utility. This study assesses whether a convolutional neural network (CNN) developed using still images and short video sequences from a multicentre dataset using different processors generalises effectively to locate polyps in a new video dataset consisting of complete colonoscopy withdrawals (caecum to rectum). Methods Our group previously developed a CNN using 4664 polyp test frames from the MICCAI 2015 polyp dataset. Here, we created a new dataset using 17 complete colonoscopy withdrawal videos, previously unseen by the CNN, containing 83 polyps consisting of 83,716 frames (14,634 polyp & 69,082 non-polyp) using Olympus EVIS LUCERA CV290(SL) processors and colonoscopes. White light frames were annotated by drawing bounding boxes around polyps. Size, morphology, histopathology and location was recorded for each polyp (table 1). Low quality frames (e.g. blurred) were excluded. Half the procedures were randomly selected to create a testing set. A true positive was scored if the CNN prediction overlapped with the bounding box. A false positive indicated a non-overlapping location. Results The CNN operated at real-time video-rate achieving a sensitivity of 91.6% and positive predictive value 75.3% in the MICCAI test set. When the MICCAI trained CNN was tested on our previously unseen colonoscopy procedures, it achieved a sensitivity of 76.6% and specificity of 78.9%. This CNN was fine-tuned by using polyp positive frames from our training dataset. This improved sensitivity to 84.5% and specificity to 92.5%. Conclusion Whilst the CNN achieved excellent results on the public still image dataset, it is more challenging to generalise results to complete colonoscopy withdrawals. Fine-tuning using our dataset improved performance. Furthermore, our procedures were performed by experts, including a significant proportion of right sided flat elevated and subtle sessile serrated lesions which were not evaluated in recent publications. AI development should include complete colonoscopy withdrawals to reflect true clinical practice and focus specifically on identifying challenging polyps that may be overlooked." @default.
• W2968146089 created "2019-08-22" @default.
• W2968146089 creator A5028279799 @default.
• W2968146089 creator A5028617313 @default.
• W2968146089 creator A5035167757 @default.
• W2968146089 creator A5042695837 @default.
• W2968146089 creator A5048400975 @default.
• W2968146089 creator A5051955250 @default.
• W2968146089 creator A5056678728 @default.
• W2968146089 creator A5060644314 @default.
• W2968146089 creator A5077630267 @default.
• W2968146089 creator A5079464849 @default.
• W2968146089 creator A5084284002 @default.
• W2968146089 date "2019-06-01" @default.
• W2968146089 modified "2023-09-27" @default.
• W2968146089 title "OTU-04 Artificial intelligence for real-time polyp localisation in colonoscopy withdrawal videos" @default.
• W2968146089 doi "https://doi.org/10.1136/gutjnl-2019-bsgabstracts.4" @default.
• W2968146089 hasPublicationYear "2019" @default.
• W2968146089 type Work @default.
• W2968146089 sameAs 2968146089 @default.
• W2968146089 citedByCount "0" @default.
• W2968146089 crossrefType "proceedings-article" @default.
• W2968146089 hasAuthorship W2968146089A5028279799 @default.
• W2968146089 hasAuthorship W2968146089A5028617313 @default.
• W2968146089 hasAuthorship W2968146089A5035167757 @default.
• W2968146089 hasAuthorship W2968146089A5042695837 @default.
• W2968146089 hasAuthorship W2968146089A5048400975 @default.
• W2968146089 hasAuthorship W2968146089A5051955250 @default.
• W2968146089 hasAuthorship W2968146089A5056678728 @default.
• W2968146089 hasAuthorship W2968146089A5060644314 @default.
• W2968146089 hasAuthorship W2968146089A5077630267 @default.
• W2968146089 hasAuthorship W2968146089A5079464849 @default.
• W2968146089 hasAuthorship W2968146089A5084284002 @default.
• W2968146089 hasBestOaLocation W29681460891 @default.
• W2968146089 hasConcept C121608353 @default.
• W2968146089 hasConcept C126322002 @default.
• W2968146089 hasConcept C153180895 @default.
• W2968146089 hasConcept C154945302 @default.
• W2968146089 hasConcept C169903167 @default.
• W2968146089 hasConcept C2778435480 @default.
• W2968146089 hasConcept C31972630 @default.
• W2968146089 hasConcept C41008148 @default.
• W2968146089 hasConcept C526805850 @default.
• W2968146089 hasConcept C71924100 @default.
• W2968146089 hasConcept C81363708 @default.
• W2968146089 hasConceptScore W2968146089C121608353 @default.
• W2968146089 hasConceptScore W2968146089C126322002 @default.
• W2968146089 hasConceptScore W2968146089C153180895 @default.
• W2968146089 hasConceptScore W2968146089C154945302 @default.
• W2968146089 hasConceptScore W2968146089C169903167 @default.
• W2968146089 hasConceptScore W2968146089C2778435480 @default.
• W2968146089 hasConceptScore W2968146089C31972630 @default.
• W2968146089 hasConceptScore W2968146089C41008148 @default.
• W2968146089 hasConceptScore W2968146089C526805850 @default.
• W2968146089 hasConceptScore W2968146089C71924100 @default.
• W2968146089 hasConceptScore W2968146089C81363708 @default.
• W2968146089 hasLocation W29681460891 @default.
• W2968146089 hasOpenAccess W2968146089 @default.
• W2968146089 hasPrimaryLocation W29681460891 @default.
• W2968146089 hasRelatedWork W10793771 @default.
• W2968146089 hasRelatedWork W11383935 @default.
• W2968146089 hasRelatedWork W12793662 @default.
• W2968146089 hasRelatedWork W1602910 @default.
• W2968146089 hasRelatedWork W2988963 @default.
• W2968146089 hasRelatedWork W40380 @default.
• W2968146089 hasRelatedWork W5386269 @default.
• W2968146089 hasRelatedWork W9190101 @default.
• W2968146089 hasRelatedWork W9388322 @default.
• W2968146089 hasRelatedWork W9783904 @default.
• W2968146089 isParatext "false" @default.
• W2968146089 isRetracted "false" @default.
• W2968146089 magId "2968146089" @default.
• W2968146089 workType "article" @default.