Autonomous Artificial Intelligence vs Artificial Intelligence-Assisted Human Optical Diagnosis of Colorectal Polyps: A Randomized Controlled Trial.

Artificial intelligence (AI)-based optical diagnosis systems (CADx) have been developed to allow pathology prediction of colorectal polyps during colonoscopies. However, CADx systems have not yet been validated for autonomous performance. Therefore, we conducted a trial comparing autonomous AI to AI-assisted human (AI-H) optical diagnosis. We performed a randomized noninferiority trial of patients undergoing elective colonoscopies at 1 academic institution. Patients were randomized into (1) autonomous AI-based CADx optical diagnosis of diminutive polyps without human input or (2) diagnosis by endoscopists who performed optical diagnosis of diminutive polyps after seeing the real-time CADx diagnosis. The primary outcome was accuracy in optical diagnosis in both arms using pathology as the gold standard. Secondary outcomes included agreement with pathology for surveillance intervals. A total of 467 patients were randomized (238 patients/158 polyps in the autonomous AI group and 229 patients/179 polyps in the AI-H group). Accuracy for optical diagnosis was 77.2% (95% confidence interval [CI], 69.7-84.7) in the autonomous AI group and 72.1% (95% CI, 65.5-78.6) in the AI-H group (P = .86). For high-confidence diagnoses, accuracy for optical diagnosis was 77.2% (95% CI, 69.7-84.7) in the autonomous AI group and 75.5% (95% CI, 67.9-82.0) in the AI-H group. Autonomous AI had statistically significantly higher agreement with pathology-based surveillance intervals compared to AI-H (91.5% [95% CI, 86.9-96.1] vs 82.1% [95% CI, 76.5-87.7]; P = .016). Autonomous AI-based optical diagnosis exhibits noninferior accuracy to endoscopist-based diagnosis. Both autonomous AI and AI-H exhibited relatively low accuracy for optical diagnosis; however, autonomous AI achieved higher agreement with pathology-based surveillance intervals. (ClinicalTrials.gov, Number NCT05236790).

Djinbachian R ，Haumesser C ，Taghiakbari M ，Pohl H ，Barkun A ，Sidani S ，Liu Chen Kiow J ，Panzini B ，Bouchard S ，Deslandres E ，Alj A ，von Renteln D ... -  《-》

Artificial Intelligence Allows Leaving-In-Situ Colorectal Polyps.

Artificial Intelligence (AI) could support cost-saving strategies for colonoscopy because of its accuracy in the optical diagnosis of colorectal polyps. However, AI must meet predefined criteria to be implemented in clinical settings. An approved computer-aided diagnosis (CADx) module for differentiating between adenoma and nonadenoma in unmagnified white-light colonoscopy was used in a consecutive series of colonoscopies. For each polyp, CADx output and subsequent endoscopist diagnosis with advanced imaging were matched against the histology gold standard. The primary outcome was the negative predictive value (NPV) of CADx for adenomatous histology for ≤5-mm rectosigmoid lesions. We also calculated the NPV for AI-assisted endoscopist predictions, and agreement between CADx and histology-based postpolypectomy surveillance intervals according to European and American guidelines. Overall, 544 polyps were removed in 162 patients, of which 295 (54.2%) were ≤5-mm rectosigmoid histologically verified lesions. CADx diagnosis was feasible in 291 of 295 (98.6%), and the NPV for ≤5-mm rectosigmoid lesions was 97.6% (95% CI, 94.1%-99.1%). There were 242 of 295 (82%) lesions that were amenable for a leave-in-situ strategy. Based on CADx output, 212 of 544 (39%) would be amenable to a resect-and-discard strategy, resulting in a 95.6% (95% CI, 90.8%-98.0%) and 95.9% (95% CI, 89.8%-98.4%) agreement between CADx- and histology-based surveillance intervals according to European and American guidelines, respectively. A similar NPV (97.6%; 95% CI, 94.8%-99.1%) for ≤5-mm rectosigmoids was achieved by AI-assisted endoscopists assessing polyps with electronic chromoendoscopy, with a CADx-concordant diagnosis in 97.2% of cases. In this study, CADx without advanced imaging exceeded the benchmarks required for optical diagnosis of colorectal polyps. CADx could help implement cost-saving strategies in colonoscopy by reducing the burden of polypectomy and/or pathology. gov registration number: NCT04884581.

Hassan C ，Balsamo G ，Lorenzetti R ，Zullo A ，Antonelli G ... -  《-》

Computer-aided diagnosis for the resect-and-discard strategy for colorectal polyps: a systematic review and meta-analysis.

The resect-and-discard strategy allows endoscopists to replace post-polypectomy pathology with real-time prediction of polyp histology during colonoscopy (optical diagnosis). We aimed to investigate the benefits and harms of implementing computer-aided diagnosis (CADx) for polyp pathology into the resect-and-discard strategy. In this systematic review and meta-analysis, we searched MEDLINE, Embase, and Scopus from database inception to June 5, 2024, without language restrictions, for diagnostic accuracy studies that assessed the performance of real-time CADx systems, compared with histology, for the optical diagnosis of diminutive polyps (≤5 mm) in the entire colon. We synthesised data for three strategies: CADx-alone, CADx-unassisted, and CADx-assisted; when the endoscopist was involved in the optical diagnosis, we synthesised data exclusively from diagnoses for which confidence in the prediction was reported as high. The primary outcomes were the proportion of polyps that would have avoided pathological assessment (ie, the proportion optically diagnosed with high confidence; main benefit) and the proportion of polyps incorrectly predicted due to false positives and false negatives (main harm), directly compared between CADx-assisted and CADx-unassisted strategies. We used DerSimonian and Laird's random-effects model to calculate all outcomes. We used Higgins I2 to assess heterogeneity, the Grading of Recommendations, Assessment, Development, and Evaluation approach to rate certainty, and funnel plots and Egger's test to examine publication bias. This study is registered with PROSPERO, CRD42024508440. We found 1019 studies, of which 11 (7400 diminutive polyps, 3769 patients, and 185 endoscopists) were included in the final meta-analysis. Three studies (1817 patients and 4086 polyps [2148 neoplastic and 1938 non-neoplastic]) provided data to directly compare the primary outcome measures between the CADx-unassisted and CADx-assisted strategies. We found no significant difference between the CADx-assisted and CADx-unassisted strategies for the proportion of polyps that would have avoided pathological assessment (90% [88-93], 3653 [89·4%] of 4086 polyps diagnosed with high confidence vs 90% [95% CI 85-94], 3588 [87·8%] of 4086 polyps diagnosed with high confidence; risk ratio 1·01 [95% CI 0·99-1·04; I2=53·49%; low-certainty evidence; Egger's test p=0·18). The proportion of incorrectly predicted polyps was lower with the CADx-assisted strategy than with the CADx-unassisted strategy (12% [95% CI 7-17], 523 [14·3%] of 3653 polyps incorrectly predicted with a CADx-assisted strategy vs 13% [6-20], 582 [16·2%] of 3588 polyps incorrectly diagnosed with a CADx-unassisted strategy; risk ratio 0·88 [95% CI 0·79-0·98]; I2=0·00%; low-certainty evidence; Egger's test p=0·18). CADx did not produce benefit nor harm for the resect-and-discard strategy, questioning its value in clinical practice. Improving the accuracy and explainability of CADx is desired. European Commission (Horizon Europe), the Japan Society of Promotion of Science, and Associazione Italiana per la Ricerca sul Cancro.

Hassan C ，Rizkala T ，Mori Y ，Spadaccini M ，Misawa M ，Antonelli G ，Rondonotti E ，Dekker E ，Houwen BBSL ，Pech O ，Baumer S ，Li JW ，von Renteln D ，Haumesser C ，Maselli R ，Facciorusso A ，Correale L ，Menini M ，Schilirò A ，Khalaf K ，Patel H ，Radadiya DK ，Bhandari P ，Kudo SE ，Sultan S ，Vandvik PO ，Sharma P ，Rex DK ，Foroutan F ，Repici A ，CADx analysis study group ... -  《The Lancet Gastroenterology & Hepatology》

Real-Time Artificial Intelligence-Based Optical Diagnosis of Neoplastic Polyps during Colonoscopy.

Barua I ，Wieszczy P ，Kudo SE ，Misawa M ，Holme Ø ，Gulati S ，Williams S ，Mori K ，Itoh H ，Takishima K ，Mochizuki K ，Miyata Y ，Mochida K ，Akimoto Y ，Kuroki T ，Morita Y ，Shiina O ，Kato S ，Nemoto T ，Hayee B ，Patel M ，Gunasingam N ，Kent A ，Emmanuel A ，Munck C ，Nilsen JA ，Hvattum SA ，Frigstad SO ，Tandberg P ，Løberg M ，Kalager M ，Haji A ，Bretthauer M ，Mori Y ... -  《-》

Artificial Intelligence for Real-Time Prediction of the Histology of Colorectal Polyps by General Endoscopists.

Rex DK ，Bhavsar-Burke I ，Buckles D ，Burton J ，Cartee A ，Comar K ，Edwards A ，Fennimore B ，Fischer M ，Gerich M ，Gilmore A ，Hamdeh S ，Hoffman J ，Ibach M ，Jackson M ，James-Stevenson T ，Kaltenbach T ，Kaplan J ，Kapur S ，Kohm D ，Kriss M ，Kundumadam S ，Kyanam Kabir Baig KR ，Menard-Katcher P ，Kraft C ，Langworthy J ，Misra B ，Molloy E ，Munoz JC ，Norvell J ，Nowak T ，Obaitan I ，Patel S ，Patel M ，Peter S ，Reid BM ，Rogers N ，Ross J ，Ryan J ，Sagi S ，Saito A ，Samo S ，Sarkis F ，Scott FI ，Siwiec R ，Sullivan S ，Wieland A ，Zhang J ，Repici A ，Hassan C ，Byrne MF ，Rastogi A ... -  《-》